Existing law provides that everyone is responsible not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person.

Existing law requires the developer of a generative artificial intelligence system or service that is released on or after January 1, 2022, and made publicly available to Californians for use, to post on the developer's internet website documentation regarding the data used by the developer to train the generative artificial intelligence system or service. Existing law defines artificial intelligence for these purposes.

This bill would prohibit a defendant who developed, modified, or used artificial intelligence, as defined, from asserting a defense that the artificial intelligence autonomously caused the harm to the plaintiff.

Existing law provides that everyone is responsible not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person.

Existing law requires the developer of a generative artificial intelligence system or service that is released on or after January 1, 2022, and made publicly available to Californians for use, to post on the developer's internet website documentation regarding the data used by the developer to train the generative artificial intelligence system or service. Existing law defines artificial intelligence for these purposes.

This bill would prohibit a defendant who developed, modified, or used artificial intelligence, as defined, from asserting a defense that the artificial intelligence autonomously caused the harm to the plaintiff.

Existing law provides that everyone is responsible not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person.

Existing law requires the developer of a generative artificial intelligence system or service that is released on or after January 1, 2022, and made publicly available to Californians for use, to post on the developer's internet website documentation regarding the data used by the developer to train the generative artificial intelligence system or service. Existing law defines artificial intelligence for these purposes.

This bill would prohibit a defendant who developed, modified, or used artificial intelligence, as defined, from asserting a defense that the artificial intelligence autonomously caused the harm to the plaintiff.

Existing law provides that everyone is responsible not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person.

Existing law requires the developer of a generative artificial intelligence system or service that is released on or after January 1, 2022, and made publicly available to Californians for use, to post on the developer's internet website documentation regarding the data used by the developer to train the generative artificial intelligence system or service. Existing law defines artificial intelligence for these purposes.

This bill would prohibit a defendant that developed who developed, modified, or used artificial intelligence, as defined, from asserting a defense that the artificial intelligence autonomously caused the harm to the plaintiff.

Existing law provides that everyone is responsible not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person.

Existing law requires the developer of a generative artificial intelligence system or service that is released on or after January 1, 2022, and made publicly available to Californians for use, to post on the developer's internet website documentation regarding the data used by the developer to train the generative artificial intelligence system or service. Existing law defines artificial intelligence for these purposes.

This bill would prohibit a defendant that developed or used artificial intelligence, as defined, from asserting a defense that the artificial intelligence autonomously caused the harm to the plaintiff.

CONCURRENCE IN SENATE AMENDMENTS

CSA1 Bill Id:AB 316

Author:(Krell)

As Amended Ver:September 2, 2025

Majority vote

SUMMARY

Establishes that in civil actions, where a plaintiff alleges harm caused by AI, a defendant who developed, modified, or used the AI is prohibited from asserting that the AI acted autonomously as a defense.

Major Provisions

1. Establishes that in civil actions, where a plaintiff alleges harm caused by AI, a defendant who developed, modified, or used the AI is prohibited from asserting that the AI acted autonomously as a defense.
2. Defines "artificial intelligence" as "engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments."
3. Clarifies that the bill does not limit or preclude a defendant from presenting any other affirmative defense, including evidence relevant to causation or foreseeability, or other evidence relevant to the comparative fault of any other person or entity.

Senate Amendments

Clarify that the bill does not limit or preclude a defendant from presenting any other affirmative defense or other evidence relevant to comparative fault.

COMMENTS

The increasing integration of artificial intelligence—systems that, with varying degrees of autonomy, generate outputs influencing physical or virtual environments—prompts a reevaluation of liability when these systems cause harm. Under current tort law, responsibility hinges on human action, with defenses available when harm stems from independent causes. AI's autonomy complicates this, as its decisions may not directly trace to human oversight, potentially allowing developers or users to avoid liability. Right now, if an AI system misdiagnoses a patient or generates defamatory content, defendants might argue the system's independence absolves them—an argument this bill seeks to eliminate.

California's tort system, rooted in Civil Code Section 1714, imposes a duty of care on all persons to avoid injuring others through negligence or willful acts, a principle adaptable to any tort claim—negligence-based (e.g., defamation, medical malpractice) or strict liability-based (e.g., products liability)—while allowing defenses like intervening causation when independent forces break the causal chain (Restatement (Second) of Torts Section 441). AI's autonomous capabilities strain this system, as its outputs may appear detached from direct human control, raising questions about accountability when harm occurs. Courts have previously adapted traditional principles to technology-related harms, such as software defects, but no statute explicitly governs AI autonomy as a defense, leaving judicial interpretation to bridge the gap.

Under California law, negligence is established when a plaintiff proves that the defendant owed a legal duty of care, breached that duty, and that the breach was both the actual and proximate cause of the plaintiff's injury, resulting in damages. (Ladd v. County of San Mateo (1996) 12 Cal.4th 913, 917; Civil Code Section 1714 (a).) Applied to AI, this framework holds developers or users liable if their oversight or deployment of the system falls below a reasonable standard of care.

For example, consider XYZ Corp., a California-based software company, which develops an AI-powered navigation system used in autonomous vehicles. The company markets the software as highly reliable, claiming it has undergone extensive testing. However, due to cost-cutting measures, XYZ Corp. skips several recommended safety protocols and releases the software without adequately testing how it responds to construction zones. When an autonomous vehicle using XYZ's software encounters an unmarked road closure, the AI misinterprets the data and accelerates instead of stopping, causing a collision that severely injures a pedestrian.

Under California's negligence standard, the injured pedestrian could prevail against XYZ Corp. for damages by proving all of the required elements of negligence:

1. Duty of care. As a software developer selling AI-based navigation technology, the corporation owes a duty of care to consumers and the general public to design and maintain a reasonably safe product. (Civil Code Section1714 (a); Rowland v. Christian (1968) 69 Cal.2d 108, 112-13.)
2. Breach of duty. XYZ Corp. breached its duty by failing to properly test its AI system under foreseeable real-world conditions, such as road closures, before deployment. A reasonable software company would have conducted rigorous testing to prevent foreseeable risks. (Mosley v. Arden Farms Co. (1945) 26 Cal.2d 213, 216.)
3. Causation. XYZ's breach was a substantial factor in causing the harm. (Mitchell v. Gonzales (1991) 54 Cal.3d 1041, 1052.) Here, the accident would not have occurred but for XYZ's defective software, making XYZ's negligence the actual cause. Additionally, the harm was foreseeable because navigation software must account for common road conditions like closures, establishing proximate cause.
4. Damages. The pedestrian suffered severe injuries, resulting in medical expenses, lost wages, and pain and suffering, which are compensable under California law.
5. Defenses and comparative negligence. XYZ Corp. may argue that the vehicle manufacturer also contributed to the harm by failing to include a manual override system. Under California's pure comparative negligence rule (Li v. Yellow Cab Co. (1975) 13 Cal.3d 804, 828-29), the pedestrian's recovery would be reduced in proportion to any fault assigned to other parties but would not be entirely barred, unless the AI's autonomy is deemed an independent cause—a defense courts currently weigh on a case-by-case.

This example shows how the existing negligence framework could address AI harm by focusing on human oversight, but AI's autonomy might complicate causation, prompting defendants to argue it acted unforeseeably, a contention this bill seeks to foreclose.

This bill is largely preemptive. It seeks to eliminate autonomy as a shield, streamlining causation across torts like defamation, malpractice, and products liability. As noted above, this "autonomy defense" has not been legally tested. Defendants facing liability for AI-related harm have instead challenged negligence and strict liability in other ways, including by raising a First Amendment defense, Section 230 preemption, and failure to meet the elements of negligence/strict liability, depending on the case. This bill merely eliminates one other possible avenue of defense. By barring autonomy defenses before they are tested in court, this bill ensures AI outputs are legally tied to developers or users, preventing them from pointing the finger at AI's so-called autonomous conduct as an "independent cause." For example, in the XYZ Corp. navigation hypothetical above, under AB 316, the pedestrian's claim would succeed if XYZ's design or testing flaws are shown, regardless of the AI's "decision" to accelerate. This bill strengthens California's tort framework for AI harm but doesn't fully resolve its complexities, marking a bold yet incomplete step in the evolving AI accountability debate.

According to the Author

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law.

Arguments in Support

The Organization for Social Media Safety explains its support for AB 316:

Given both the alarming speed at which AI-based products are being deployed and the clear, convincing proof that these products can cause severe harm, especially to children, we must ensure that our standard product liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California's products have an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

Your AB 316 (Krell) does just that. It codifies a straightforward principle: those who profit from the development and use of AI cannot evade liability by blaming the intended autonomy of their own machines. If Big Social wishes to profit from this technology, it must do so within the framework of existing tort law, which incentivizes the consideration of safety in product development and deployment. In this way, we can protect against more broken children.

The University of San Diego's Children's Advocacy Institute explains the urgency of this measure, against a speedily changing backdrop:

[W]e cannot wait for the most basic legal question about autonomously operating AI – who is legally responsible if it harms someone -- to be clarified definitively through decades of litigation. It took 25 years for courts finally to decree that Amazon's marketplace was subject to the same products liability laws as a corner store selling the identical product. It has taken 20 years for courts to figure out how to apply basic negligence and public nuisance law to the harms caused by these companies; a process that is still evolving. According to the most recent data available, the median time frame for resolving just the appeal of a case in California (not including years for the trial) is about a year and a half. That's the median; by definition half of the appeals take longer.

Hence, AB 316 (Krell) simply codifies what should be obvious: that the persons who profit from the development and use of AI cannot, if the AI harms someone or violates a law, point a blaming finger at the intended autonomy of their own machine in an effort to escape liability.

Arguments in Opposition

TechNet, a network of CEOs and senior executives from tech industry, raises the following concerns:

We are concerned about strict liability that this bill would create. The bill disallows the defense that "artificial autonomously caused the harm." It eliminates key arguments regarding foreseeability, design diligence, and user error from the conversation. This would risk imposing a strict liability for any harm allegedly associated with AI and open the door to frivolous lawsuits claiming the alleged fraud. Additionally, the bill does no demonstrate how courts should evaluate complex questions regarding causation, risk foreseeability, or comparative fault when multiple parties are involved in an AI system as developers and users are included in the same liability. As industry and policymakers alike continue to pursue the responsible deployment of AI systems, it is necessary to recognize that the technology will evolve considerably over time. There are multiple, ongoing efforts within the industry as well as national entities to leverage safety tools and create national and international transparency standards to create a cohesive framework that prioritizes responsible AI deployment. Companies are implementing safety frameworks and developing AI-based detection technologies to monitor other AI systems.

FISCAL COMMENTS

According to the Senate Appropriations Committee:

1. Unknown, potentially significant costs to the state funded trial court system (Trial Court Trust Fund, General Fund). The elimination of a defense, as proposed by this bill, may increase the likelihood of success for plaintiffs and encourage plaintiffs to file suits that they otherwise would not have. The fiscal impact of this bill to the courts will depend on many unknowns, including the number of cases filed and the factors unique to each case. An eight-hour court day costs approximately $10,500 in staff in workload. If court days exceed 10, costs to the trial courts could reach hundreds of thousands of dollars. While the courts are not funded on a workload basis, an increase in workload could result in delayed court services and would put pressure on the General Fund to fund additional staff and resources and to increase the amount appropriated to backfill for trial court operations.
2. Unknown, potentially significant costs (General Fund, local funds) to state and local agencies for increased exposure to civil liability to the extent they use artificial intelligence. Agencies may also incur higher liability insurance costs because of increased litigation exposure.

VOTES:

ASM JUDICIARY: 12-0-0

YES: Kalra, Dixon, Bauer-Kahan, Bryan, Connolly, Essayli, Harabedian, Pacheco, Papan, Sanchez, Stefani, Zbur

ASM PRIVACY AND CONSUMER PROTECTION: 11-2-2

YES: Bauer-Kahan, Dixon, Bryan, DeMaio, Irwin, Macedo, Ortega, Patterson, Petrie-Norris, Ward, Wilson

NO: Lowenthal, McKinnor

ABS, ABST OR NV: Pellerin, Wicks

ASSEMBLY FLOOR: 70-1-8

YES: Addis, Aguiar-Curry, Ahrens, Alanis, Alvarez, Arambula, Ávila Farías, Bauer-Kahan, Bennett, Berman, Boerner, Bonta, Bryan, Calderon, Caloza, Carrillo, Castillo, Chen, Connolly, Davies, DeMaio, Elhawary, Ellis, Fong, Gabriel, Gallagher, Garcia, Gipson, Mark González, Hadwick, Haney, Harabedian, Hart, Hoover, Irwin, Jackson, Kalra, Krell, Lackey, Lee, Lowenthal, Macedo, Muratsuchi, Nguyen, Ortega, Pacheco, Patel, Patterson, Petrie-Norris, Quirk-Silva, Ramos, Ransom, Celeste Rodriguez, Michelle Rodriguez, Rogers, Blanca Rubio, Sanchez, Schiavo, Schultz, Sharp-Collins, Solache, Soria, Stefani, Ta, Wallis, Ward, Wicks, Wilson, Zbur, Rivas

NO: McKinnor

ABS, ABST OR NV: Bains, Dixon, Flora, Jeff Gonzalez, Papan, Pellerin, Tangipa, Valencia

UPDATED

VERSION: September 2, 2025

CONSULTANT: Shiran Zohar / JUD. / (916) 319-2334 FN: 0001701

 THIRD READING  

Bill No: AB 316

Author: Krell (D)

Amended: 9/2/25 in Senate

Vote: 21

SENATE JUDICIARY COMMITTEE: 12-0, 6/24/25

AYES: Umberg, Niello, Allen, Arreguín, Ashby, Caballero, Durazo, Laird, Stern, Wahab, Weber Pierson, Wiener

NO VOTE RECORDED: Valladares

SENATE APPROPRIATIONS COMMITTEE: 7-0, 8/29/25

AYES: Caballero, Seyarto, Cabaldon, Dahle, Grayson, Richardson, Wahab

ASSEMBLY FLOOR: 70-1, 5/19/25 - See last page for vote

SUBJECT: Artificial intelligence: defenses

SOURCE: Children’s Advocacy Institute

Organization for Social Media Safety

DIGEST:

This bill prohibits a defendant from asserting the defense that artificial intelligence (AI) autonomously caused harm to a plaintiff, as provided.

Senate Floor amendments of 9/2/25 clarify the bill’s impact on presenting certain defenses and evidence.

ANALYSIS:

Existing law:

1. Provides that every person is responsible, not only for the result of their willful acts, but also for an injury occasioned to another by the person’s want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civil (Civ.) Code § 1714(a).)
2. Defines “artificial intelligence” as an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments. (Gov. Code § 11546.45.5.)

This bill:

1. Provides that in an action against a defendant who developed, modified, or used AI that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.
2. Clarifies that this does not limit or preclude a defendant from presenting any other affirmative defense or other relevant evidence of comparative fault.

Background

As AI models and applications become more sophisticated and integrated into our daily lives, they introduce new safety and security risks. Automated systems can make critical errors in high-stakes situations like self-driving vehicles, medical diagnostics, or home security systems when they encounter edge cases or adversarial inputs. AI-powered chatbots, phishing, identity theft, and deepfakes create novel threats to personal security and assets. Additionally, over-reliance on AI systems without adequate human oversight in critical infrastructure or emergency response could lead to cascading failures during unusual circumstances. While these technologies offer tremendous benefits, ensuring the highest level of due care on the part of AI developers and deployers is of paramount importance. Generally, individuals and entities are not only liable for their willful acts but also for injuries caused by their lack of ordinary care in managing their property or person. However, there are concerns that existing legal frameworks may be challenged in addressing the unique risks and complexities of AI technologies.

This bill addresses the issue by making clear that a defendant who developed, modified, or used AI that is alleged to have caused a harm to a plaintiff, cannot assert as a defense that the AI autonomously caused the harm to the plaintiff. This bill is sponsored by the Children’s Advocacy Institute and the Organization for Social Media Safety. It is supported by a number of organizations, including the California Federation of Labor Unions and the California Initiative for Technology and Democracy (CITED). This bill is opposed by Technet and the California Chamber of Commerce.

Comments

Negligence law serves a crucial purpose in our legal system by incentivizing individuals and companies to take reasonable precautions to prevent harm. When organizations face potential liability for negligent design, testing, or deployment, they are motivated to invest in robust safety measures, thorough testing protocols, and ongoing risk monitoring. When a company is held responsible for harms it causes through its systems or products, it prompts a proactive approach to avoid causing those harms. This creates a direct financial incentive to prioritize safety, especially when utilizing new technologies. This is the state of the law currently.

This bill makes clear that it shall not be a defense, and a defendant shall not assert, that AI developed, modified, or used by the defendant autonomously caused alleged harm to a plaintiff. This ensures that AI development and deployment is done with due care despite the novel nature of the technology and its inherent complexities. Ultimately, this preserves the principle that humans are responsible for the harms they cause, regardless of the sophistication or autonomy of the tools they use.

While there are no examples of defendants successfully utilizing such defenses, this bill proactively rules out this avenue of deflecting blame when someone suffers AI-related injuries. One example of where this defense has been attempted was recently widely reported on:

In 2022, Air Canada’s chatbot promised a discount that wasn’t available to passenger Jake Moffatt, who was assured that he could book a full-fare flight for his grandmother’s funeral and then apply for a bereavement fare after the fact.

According to a civil-resolutions tribunal decision last Wednesday, when Moffatt applied for the discount, the airline said the chatbot had been wrong – the request needed to be submitted before the flight – and it wouldn’t offer the discount. Instead, the airline said the chatbot was a “separate legal entity that is responsible for its own actions”. Air Canada argued that Moffatt should have gone to the link provided by the chatbot, where he would have seen the correct policy.

The British Columbia Civil Resolution Tribunal rejected that argument, ruling that Air Canada had to pay Moffatt $812.02 (£642.64) in damages and tribunal fees. “It should be obvious to Air Canada that it is responsible for all the information on its website,” read tribunal member Christopher Rivers’ written response. “It makes no difference whether the information comes from a static page or a chatbot.”^[1]

A recent article highlights the importance of addressing this issue and heading off any legal doctrine that allows AI itself to be blamed:

On the question of any eventual push to juridical personhood for AIs, it does seem that a cautionary note is needed. Certainly, it can be strongly argued in terms both of equity and deterrence, that there is, in the present state of things, a need to avoid ideas or developments that may see those persons and corporate entities that have profited from the development of AIs ultimately avoid liability for the consequences of their creation, particularly where those consequences are harmful.^[2]

According to the author:

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law.

FISCAL EFFECT: Appropriation: No Fiscal Com.: No Local: No

According to the Senate Appropriations Committee:

• Unknown, potentially significant costs to the state funded trial court system (Trial Court Trust Fund, General Fund). The elimination of a defense, as proposed by this bill, may increase the likelihood of success for plaintiffs and encourage plaintiffs to file suits that they otherwise would not have. The fiscal impact of this bill to the courts will depend on many unknowns, including the number of cases filed and the factors unique to each case. An eight-hour court day costs approximately $10,500 in staff in workload. If court days exceed 10, costs to the trial courts could reach hundreds of thousands of dollars. While the courts are not funded on a workload basis, an increase in workload could result in delayed court services and would put pressure on the General Fund to fund additional staff and resources and to increase the amount appropriated to backfill for trial court operations.
• Unknown, potentially significant costs (General Fund, local funds) to state and local agencies for increased exposure to civil liability to the extent they use artificial intelligence. Agencies may also incur higher liability insurance costs because of increased litigation exposure.

SUPPORT: (Verified 9/2/25)

Children’s Advocacy Institute (source)

Organization for Social Media Safety (source)

3strands Global Foundation

California Civil Liberties Advocacy

California Conference Board of the Amalgamated Transit Union

California Conference of Machinists

California Federation of Labor Unions, AFL-CIO

California Initiative for Technology & Democracy

California Nurses Association

California Teamsters Public Affairs Council

Consumer Attorneys of California

Consumer Federation of California

Economic Security California Action

Engineers and Scientists of California, IFPTE Local 20, AFL-CIO

Oakland Privacy

Tech Oversight California

TechEquity Action

The Center for AI and Digital Policy

UFCW - Western States Council

UNITE Here International Union, AFL-CIO

Utility Workers Union of America

OPPOSITION: (Verified 9/2/25)

California Chamber of Commerce

Technet

ARGUMENTS IN SUPPORT: The Organization for Social Media Safety, a sponsor of this bill, writes:

Given both the alarming speed at which AI-based tools are being deployed and the clear, convincing proof that these tools can cause severe harm, especially to children, we must ensure that our standard liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California’s marketplace has an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

ARGUMENTS IN OPPOSITION: Technet and the California Chamber of Commerce write in joint opposition:

We understand the intent to prevent defendants from attempting to absolve themselves from liability by claiming that an artificial intelligence acted autonomously. Our concern is that the bill could be interpreted to prevent a defendant from presenting any evidence related to an AI or automated system, which may be relevant to causation, foreseeability of harm, and the comparative fault of other parties.

ASSEMBLY FLOOR: 70-1, 5/19/25

AYES: Addis, Aguiar-Curry, Ahrens, Alanis, Alvarez, Arambula, Ávila Farías, Bauer-Kahan, Bennett, Berman, Boerner, Bonta, Bryan, Calderon, Caloza, Carrillo, Castillo, Chen, Connolly, Davies, DeMaio, Elhawary, Ellis, Fong, Gabriel, Gallagher, Garcia, Gipson, Mark González, Hadwick, Haney, Harabedian, Hart, Hoover, Irwin, Jackson, Kalra, Krell, Lackey, Lee, Lowenthal, Macedo, Muratsuchi, Nguyen, Ortega, Pacheco, Patel, Patterson, Petrie-Norris, Quirk-Silva, Ramos, Ransom, Celeste Rodriguez, Michelle Rodriguez, Rogers, Blanca Rubio, Sanchez, Schiavo, Schultz, Sharp-Collins, Solache, Soria, Stefani, Ta, Wallis, Ward, Wicks, Wilson, Zbur, Rivas

NOES: McKinnor

NO VOTE RECORDED: Bains, Dixon, Flora, Jeff Gonzalez, Papan, Pellerin, Tangipa, Valencia

Prepared by: Christian Kurpiewski / JUD. / (916) 651-4113

9/3/25 18:31:21

**** END ****

1. Maria Yagoda, Airline held liable for its chatbot giving passenger bad advice - what this means for travelers (February 23, 2024) BBC, https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know. ↑

2. Michael Duffy, Rise of the ‘Machine Defendant’? A Cautionary Analysis and Conceptualisation of Civil and Criminal Liability Approaches to the Actions of Robots and Artificial Intelligence (January 1, 2023). 49(2) Monash University Law Review 1-42, http://dx.doi.org/10.2139/ssrn.5032505. ↑

 THIRD READING  

Bill No: AB 316

Author: Krell (D)

Amended: 4/28/25 in Assembly

Vote: 21

SENATE JUDICIARY COMMITTEE: 12-0, 6/24/25

AYES: Umberg, Niello, Allen, Arreguín, Ashby, Caballero, Durazo, Laird, Stern, Wahab, Weber Pierson, Wiener

NO VOTE RECORDED: Valladares

SENATE APPROPRIATIONS COMMITTEE: 7-0, 8/29/25

AYES: Caballero, Seyarto, Cabaldon, Dahle, Grayson, Richardson, Wahab

ASSEMBLY FLOOR: 70-1, 5/19/25 - See last page for vote

SUBJECT: Artificial intelligence: defenses

SOURCE: Children’s Advocacy Institute

Organization for Social Media Safety

DIGEST:

This bill prohibits a defendant from asserting the defense that artificial intelligence (AI) autonomously caused harm to a plaintiff, as provided.

ANALYSIS:

Existing law:

1. Provides that every person is responsible, not only for the result of their willful acts, but also for an injury occasioned to another by the person’s want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civil (Civ.) Code § 1714(a).)
2. Defines “artificial intelligence” as an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments. (Gov. Code § 11546.45.5.)

This bill provides that in an action against a defendant who developed, modified, or used AI that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.

Background

As AI models and applications become more sophisticated and integrated into our daily lives, they introduce new safety and security risks. Automated systems can make critical errors in high-stakes situations like self-driving vehicles, medical diagnostics, or home security systems when they encounter edge cases or adversarial inputs. AI-powered chatbots, phishing, identity theft, and deepfakes create novel threats to personal security and assets. Additionally, over-reliance on AI systems without adequate human oversight in critical infrastructure or emergency response could lead to cascading failures during unusual circumstances. While these technologies offer tremendous benefits, ensuring the highest level of due care on the part of AI developers and deployers is of paramount importance. Generally, individuals and entities are not only liable for their willful acts but also for injuries caused by their lack of ordinary care in managing their property or person. However, there are concerns that existing legal frameworks may be challenged in addressing the unique risks and complexities of AI technologies.

This bill addresses the issue by making clear that a defendant who developed, modified, or used AI that is alleged to have caused a harm to a plaintiff, cannot assert as a defense that the AI autonomously caused the harm to the plaintiff. This bill is sponsored by the Children’s Advocacy Institute and the Organization for Social Media Safety. It is supported by a number of organizations, including the California Federation of Labor Unions and the California Initiative for Technology and Democracy (CITED). This bill is opposed by Technet and the California Chamber of Commerce.

Comment

Negligence law serves a crucial purpose in our legal system by incentivizing individuals and companies to take reasonable precautions to prevent harm. When organizations face potential liability for negligent design, testing, or deployment, they are motivated to invest in robust safety measures, thorough testing protocols, and ongoing risk monitoring. When a company is held responsible for harms it causes through its systems or products, it prompts a proactive approach to avoid causing those harms. This creates a direct financial incentive to prioritize safety, especially when utilizing new technologies. This is the state of the law currently.

This bill makes clear that it shall not be a defense, and a defendant shall not assert, that AI developed, modified, or used by the defendant autonomously caused alleged harm to a plaintiff. This ensures that AI development and deployment is done with due care despite the novel nature of the technology and its inherent complexities. Ultimately, this preserves the principle that humans are responsible for the harms they cause, regardless of the sophistication or autonomy of the tools they use.

While there are no examples of defendants successfully utilizing such defenses, this bill proactively rules out this avenue of deflecting blame when someone suffers AI-related injuries. One example of where this defense has been attempted was recently widely reported on:

In 2022, Air Canada’s chatbot promised a discount that wasn’t available to passenger Jake Moffatt, who was assured that he could book a full-fare flight for his grandmother’s funeral and then apply for a bereavement fare after the fact.

According to a civil-resolutions tribunal decision last Wednesday, when Moffatt applied for the discount, the airline said the chatbot had been wrong – the request needed to be submitted before the flight – and it wouldn’t offer the discount. Instead, the airline said the chatbot was a “separate legal entity that is responsible for its own actions”. Air Canada argued that Moffatt should have gone to the link provided by the chatbot, where he would have seen the correct policy.

The British Columbia Civil Resolution Tribunal rejected that argument, ruling that Air Canada had to pay Moffatt $812.02 (£642.64) in damages and tribunal fees. “It should be obvious to Air Canada that it is responsible for all the information on its website,” read tribunal member Christopher Rivers’ written response. “It makes no difference whether the information comes from a static page or a chatbot.”^[1]

A recent article highlights the importance of addressing this issue and heading off any legal doctrine that allows AI itself to be blamed:

On the question of any eventual push to juridical personhood for AIs, it does seem that a cautionary note is needed. Certainly, it can be strongly argued in terms both of equity and deterrence, that there is, in the present state of things, a need to avoid ideas or developments that may see those persons and corporate entities that have profited from the development of AIs ultimately avoid liability for the consequences of their creation, particularly where those consequences are harmful.^[2]

According to the author:

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law.

FISCAL EFFECT: Appropriation: No Fiscal Com.: No Local: No

According to the Senate Appropriations Committee:

• Unknown, potentially significant costs to the state funded trial court system (Trial Court Trust Fund, General Fund). The elimination of a defense, as proposed by this bill, may increase the likelihood of success for plaintiffs and encourage plaintiffs to file suits that they otherwise would not have. The fiscal impact of this bill to the courts will depend on many unknowns, including the number of cases filed and the factors unique to each case. An eight-hour court day costs approximately $10,500 in staff in workload. If court days exceed 10, costs to the trial courts could reach hundreds of thousands of dollars. While the courts are not funded on a workload basis, an increase in workload could result in delayed court services and would put pressure on the General Fund to fund additional staff and resources and to increase the amount appropriated to backfill for trial court operations.
• Unknown, potentially significant costs (General Fund, local funds) to state and local agencies for increased exposure to civil liability to the extent they use artificial intelligence. Agencies may also incur higher liability insurance costs because of increased litigation exposure.

SUPPORT: (Verified 8/29/25)

Children’s Advocacy Institute (sponsor)

Organization for Social Media Safety (sponsor)

3strands Global Foundation

California Civil Liberties Advocacy

California Conference Board of the Amalgamated Transit Union

California Conference of Machinists

California Federation of Labor Unions, AFL-CIO

California Initiative for Technology & Democracy

California Nurses Association

California Teamsters Public Affairs Council

Consumer Attorneys of California

Consumer Federation of California

Economic Security California Action

Engineers and Scientists of California, IFPTE Local 20, AFL-CIO

Oakland Privacy

Tech Oversight California

TechEquity Action

The Center for AI and Digital Policy

UFCW - Western States Council

UNITE Here International Union, AFL-CIO

Utility Workers Union of America

OPPOSITION: (Verified 8/29/25)

California Chamber of Commerce

Technet

ARGUMENTS IN SUPPORT: The Organization for Social Media Safety, a sponsor of the bill, writes:

Given both the alarming speed at which AI-based tools are being deployed and the clear, convincing proof that these tools can cause severe harm, especially to children, we must ensure that our standard liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California’s marketplace has an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

ARGUMENTS IN OPPOSITION: Technet and the California Chamber of Commerce write in joint opposition:

We understand the intent to prevent defendants from attempting to absolve themselves from liability by claiming that an artificial intelligence acted autonomously. Our concern is that the bill could be interpreted to prevent a defendant from presenting any evidence related to an AI or automated system, which may be relevant to causation, foreseeability of harm, and the comparative fault of other parties.

ASSEMBLY FLOOR: 70-1, 5/19/25

AYES: Addis, Aguiar-Curry, Ahrens, Alanis, Alvarez, Arambula, Ávila Farías, Bauer-Kahan, Bennett, Berman, Boerner, Bonta, Bryan, Calderon, Caloza, Carrillo, Castillo, Chen, Connolly, Davies, DeMaio, Elhawary, Ellis, Fong, Gabriel, Gallagher, Garcia, Gipson, Mark González, Hadwick, Haney, Harabedian, Hart, Hoover, Irwin, Jackson, Kalra, Krell, Lackey, Lee, Lowenthal, Macedo, Muratsuchi, Nguyen, Ortega, Pacheco, Patel, Patterson, Petrie-Norris, Quirk-Silva, Ramos, Ransom, Celeste Rodriguez, Michelle Rodriguez, Rogers, Blanca Rubio, Sanchez, Schiavo, Schultz, Sharp-Collins, Solache, Soria, Stefani, Ta, Wallis, Ward, Wicks, Wilson, Zbur, Rivas

NOES: McKinnor

NO VOTE RECORDED: Bains, Dixon, Flora, Jeff Gonzalez, Papan, Pellerin, Tangipa, Valencia

Prepared by: Christian Kurpiewski / JUD. / (916) 651-4113

8/30/25 11:08:58

**** END ****

1. Maria Yagoda, Airline held liable for its chatbot giving passenger bad advice - what this means for travelers (February 23, 2024) BBC, https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know. ↑

2. Michael Duffy, Rise of the ‘Machine Defendant’? A Cautionary Analysis and Conceptualisation of Civil and Criminal Liability Approaches to the Actions of Robots and Artificial Intelligence (January 1, 2023). 49(2) Monash University Law Review 1-42, http://dx.doi.org/10.2139/ssrn.5032505. ↑

SENATE JUDICIARY COMMITTEE

Senator Thomas Umberg, Chair

2025-2026 Regular Session

Bill No: AB 316 (Author:Krell)

Version: April 28, 2025

Hearing Date: June 24, 2025

Fiscal: No

Urgency: No

Consultant: CK

SUBJECT

Artificial intelligence: defenses

DIGEST

This bill prohibits a defendant from asserting the defense that AI autonomously caused harm to a plaintiff, as provided.

EXECUTIVE SUMMARY

As artificial intelligence models and applications become more sophisticated and integrated into our daily lives, they introduce new safety and security risks. Automated systems can make critical errors in high-stakes situations like self-driving vehicles, medical diagnostics, or home security systems when they encounter edge cases or adversarial inputs. AI-powered chatbots, phishing, identity theft, and deepfakes create novel threats to personal security and assets. Additionally, over-reliance on AI systems without adequate human oversight in critical infrastructure or emergency response could lead to cascading failures during unusual circumstances. While these technologies offer tremendous benefits, ensuring the highest level of due care on the part of AI developers and deployers is of paramount importance. Generally, individuals and entities are not only liable for their willful acts but also for injuries caused by their lack of ordinary care in managing their property or person. However, there are concerns that existing legal frameworks may be challenged in addressing the unique risks and complexities of AI technologies.

This bill addresses the issue by making clear that a defendant who developed, modified, or used AI that is alleged to have caused a harm to a plaintiff, cannot assert as a defense that the AI autonomously caused the harm to the plaintiff. This bill is sponsored by the Children’s Advocacy Institute and the Organization for Social Media Safety. It is supported by a number of organizations, including the California Federation of Labor Unions and the California Initiative for Technology and Democracy (CITED). The bill is opposed by Technet and the California Chamber of Commerce.

PROPOSED CHANGES TO THE LAW

Existing law:

1. Provides that every person is responsible, not only for the result of their willful acts, but also for an injury occasioned to another by the person’s want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civ. Code § 1714(a).)
2. Defines “artificial intelligence” as an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments. (Gov. Code § 11546.45.5.)

This bill provides that in an action against a defendant who developed, modified, or used AI that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.

COMMENTS

1. The risks presented by AI models and applications

With recent dramatic advances in the capabilities of AI systems, the need for frameworks for accountability and responsible development have become ever more urgent.

In January of 2017, AI researchers, economists, legal scholars, ethicists, and philosophers met in Asilomar, California, to discuss principles for managing the responsible development of AI. The collaboration resulted in the Asilomar Principles. Aspirational rather than prescriptive, these 23 principles were intended to initiate and frame a dialogue by providing direction and guidance for policymakers, researchers, and developers. The Legislature subsequently adopted ACR 215 (Kiley, Ch. 206, Stats. 2018), which added the State of California to that list by endorsing the Asilomar Principles as guiding values for the development of artificial intelligence and related public policy. One key admonition from these principles is to “recognize that [AI’s] risks are potentially catastrophic or existential.”

As directed by the National AI Initiative Act of 2020, the National Institute of Standards and Technology (NIST) developed the AI Risk Management Framework to assist entities designing, developing, deploying, and using AI systems to help manage the many risks of AI and promote trustworthy and responsible development and use of AI systems. That framework highlights the serious risks at play and the uniquely challenging nature of addressing them in this context:

Artificial intelligence (AI) technologies have significant potential to transform society and people’s lives – from commerce and health to transportation and cybersecurity to the environment and our planet. AI technologies can drive inclusive economic growth and support scientific advancements that improve the conditions of our world. AI technologies, however, also pose risks that can negatively impact individuals, groups, organizations, communities, society, the environment, and the planet. Like risks for other types of technology, AI risks can emerge in a variety of ways and can be characterized as long- or short-term, high or low-probability, systemic or localized, and high- or low-impact.

While there are myriad standards and best practices to help organizations mitigate the risks of traditional software or information-based systems, the risks posed by AI systems are in many ways unique. AI systems, for example, may be trained on data that can change over time, sometimes significantly and unexpectedly, affecting system functionality and trustworthiness in ways that are hard to understand. AI systems and the contexts in which they are deployed are frequently complex, making it difficult to detect and respond to failures when they occur. AI systems are inherently socio-technical in nature, meaning they are influenced by societal dynamics and human behavior. AI risks – and benefits – can emerge from the interplay of technical aspects combined with societal factors related to how a system is used, its interactions with other AI systems, who operates it, and the social context in which it is deployed.

These risks make AI a uniquely challenging technology to deploy and utilize both for organizations and within society. [. . .]

AI risk management is a key component of responsible development and use of AI systems. Responsible AI practices can help align the decisions about AI system design, development, and uses with intended aim and values. Core concepts in responsible AI emphasize human centricity, social responsibility, and sustainability. AI risk management can drive responsible uses and practices by prompting organizations and their internal teams who design, develop, and deploy AI to think more critically about context and potential or unexpected negative and positive impacts. Understanding and managing the risks of AI systems will help to enhance trustworthiness, and in turn, cultivate public trust.

This highlights how the risks posed by AI are inherently complex and ever-changing. Constant adaptions and nimble responses to addressing potential risks is of critical importance.

More recently the Biden Administration published its Blueprint for an AI Bill of Rights, which is a set of five principles and associated practices to help guide the design, use, and deployment of AI to protect the rights of the American public. One key piece focuses on the safety of these systems: “Safe and Effective Systems: You should be protected from unsafe or ineffective systems. Automated systems should be developed with consultation from diverse communities, stakeholders, and domain experts to identify concerns, risks, and potential impacts of the system.”^[1]

While the future is unclear, the need to respond to these potential harms now is evident. The Center for New American Security puts a fine point on it:

While there is significant uncertainty in how the future of AI develops, current trends point to a future of vastly more powerful AI systems than today’s state of the art. The most advanced systems at AI’s frontier will be limited initially to a small number of actors but may rapidly proliferate. Policymakers should begin to put in place today a regulatory framework to prepare for this future. Building an anticipatory regulatory framework is essential because of the disconnect in speeds between AI progress and the policymaking process, the difficulty in predicting the capabilities of new AI systems for specific tasks, and the speed with which AI models proliferate today, absent regulation. Waiting to regulate frontier AI systems until concrete harms materialize will almost certainly result in regulation being too late.^[2]

1. Civil liability and immunity

As a general rule, California law provides that persons are responsible, not only for the result of their willful acts, but also for an injury occasioned to another by their want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civ. Code § 1714(a).) Liability has the primary effect of ensuring that some measure of recourse exists for those persons injured by the negligent or willful acts of others; the risk of that liability has the primary effect of ensuring parties act reasonably to avoid harm to those to whom they owe a duty.

Conversely, immunity from liability disincentivizes careful planning and acting on the part of individuals and entities. When one enjoys immunity from civil liability, they are relieved of the responsibility to act with due regard and an appropriate level of care in the conduct of their activities. Immunity provisions are also disfavored because they, by their nature, preclude parties from recovering when they are injured, and force injured parties to absorb losses for which they are not responsible. Liability acts not only to allow a victim to be made whole, but to encourage appropriate compliance with legal requirements.

1. Definitively prohibiting the autonomous AI defense

Negligence law serves a crucial purpose in our legal system by incentivizing individuals and companies to take reasonable precautions to prevent harm. When organizations face potential liability for negligent design, testing, or deployment, they are motivated to invest in robust safety measures, thorough testing protocols, and ongoing risk monitoring. When a company is held responsible for harms it causes through its systems or products, it prompts a proactive approach to avoid causing those harms. This creates a direct financial incentive to prioritize safety, especially when utilizing new technologies. This is the state of the law currently.

This bill makes clear that it shall not be a defense, and a defendant shall not assert, that AI developed, modified, or used by the defendant autonomously caused alleged harm to a plaintiff. This ensures that AI development and deployment is done with due care despite the novel nature of the technology and its inherent complexities. Ultimately, this preserves the principle that humans are responsible for the harms they cause, regardless of the sophistication or autonomy of the tools they use.

While there are no examples of defendants successfully utilizing such defenses, this bill proactively rules out this avenue of deflecting blame when someone suffers AI-related injuries. One example of where this defense has been attempted was recently widely reported on:

In 2022, Air Canada’s chatbot promised a discount that wasn’t available to passenger Jake Moffatt, who was assured that he could book a full-fare flight for his grandmother’s funeral and then apply for a bereavement fare after the fact.

According to a civil-resolutions tribunal decision last Wednesday, when Moffatt applied for the discount, the airline said the chatbot had been wrong – the request needed to be submitted before the flight – and it wouldn’t offer the discount. Instead, the airline said the chatbot was a “separate legal entity that is responsible for its own actions”. Air Canada argued that Moffatt should have gone to the link provided by the chatbot, where he would have seen the correct policy.

The British Columbia Civil Resolution Tribunal rejected that argument, ruling that Air Canada had to pay Moffatt $812.02 (£642.64) in damages and tribunal fees. “It should be obvious to Air Canada that it is responsible for all the information on its website,” read tribunal member Christopher Rivers’ written response. “It makes no difference whether the information comes from a static page or a chatbot.”^[3]

A recent article highlights the importance of addressing this issue and heading off any legal doctrine that allows AI itself to be blamed:

On the question of any eventual push to juridical personhood for AIs, it does seem that a cautionary note is needed. Certainly, it can be strongly argued in terms both of equity and deterrence, that there is, in the present state of things, a need to avoid ideas or developments that may see those persons and corporate entities that have profited from the development of AIs ultimately avoid liability for the consequences of their creation, particularly where those consequences are harmful.^[4]

According to the author:

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law.

1. Stakeholder positions

The Organization for Social Media Safety, a sponsor of the bill, writes:

Given both the alarming speed at which AI-based tools are being deployed and the clear, convincing proof that these tools can cause severe harm, especially to children, we must ensure that our standard liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California’s marketplace has an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

Technet and the California Chamber of Commerce write in joint opposition:

We understand the intent to prevent defendants from attempting to absolve themselves from liability by claiming that an artificial intelligence acted autonomously. Our concern is that the bill could be interpreted to prevent a defendant from presenting any evidence related to an AI or automated system, which may be relevant to causation, foreseeability of harm, and the comparative fault of other parties.

It should be noted that this bill simply prevents defendants from asserting that AI caused the harm on its own. Plaintiffs are still required to establish all the elements of their causes of action, and defendants are still permitted to present relevant evidence to rebut that evidence.

Writing in support, CITED argues:

Artificial intelligence offers unprecedented opportunities for advances in untold fields. But this technological marvel can also be used to cause harm, both large and small. Whether it is defamatory deepfakes designed to fraudulently impact voting, chatbots that mimic human conversations and encourage children to harm themselves, price collusion among competitors designed to maximize profits at the expense of consumers, or the creation of child sexual abuse material, AI can be developed or deployed in ways that cause grave harm.

Those who reap the benefits of AI must also accept the risks and responsibilities. AB 316 (Krell) takes a small step in that direction by making clear that those who cause harm through the use of AI that they have created or deployed should not be able to avoid legal responsibility by claiming that the AI itself -- autonomously -- caused the harm.

As drafted, this bill does not in any way limit the plaintiff’s burden of proof or make it easier for them to prove their case. While we believe that California ought to lead the nation in establishing an AI liability regime so Californians who are injured by AI can appropriately seek redress in the courts, as they can with all other products and all other industries, this bill does not do that. It simply says that no one should be able to shirk responsibility for the harm they cause through a consumer’s use of their AI product with a specious legal argument about that product having its own autonomy.

SUPPORT

Children’s Advocacy Institute (sponsor)

Organization for Social Media Safety (sponsor)

3strands Global Foundation

California Conference Board of the Amalgamated Transit Union

California Conference of Machinists

California Federation of Labor Unions, AFL-CIO

California Initiative for Technology & Democracy, a Project of California Common CAUSE

California Nurses Association

California Teamsters Public Affairs Council

Consumer Attorneys of California

Consumer Federation of California

Economic Security California Action

Engineers and Scientists of California, IFPTE Local 20, AFL-CIO

Oakland Privacy

TechEquity Action

The Center for AI and Digital Policy

UFCW - Western States Council

UNITE Here International Union, AFL-CIO

Utility Workers Union of America

OPPOSITION

California Chamber of Commerce

Technet

RELATED LEGISLATION

Pending Legislation:

SB 243 (Padilla, 2025) requires operators of “companion chatbot platforms” that allow users to engage with chatbots to take reasonable steps to prevent their chatbots from engaging in specified conduct, including offering unpredictable rewards and encouraging increased engagement. Operators must periodically remind users that the chatbot is not human and implement protocols for addressing suicidal ideation expressed by users, as well as conduct annual audits. SB 243 is currently in the Assembly Privacy and Consumer Protection Committee.

SB 420 (Padilla, 2025) regulates the use of high-risk automated decision systems (ADS). This includes requirements on developers and deployers to perform impact assessments on their systems. The bill establishes the right of individuals to know when an ADS has been used, details about the systems, and an opportunity to appeal ADS decisions, where technically feasible. SB 420 is currently in the Assembly Privacy and Consumer Protection Committee.

SB 468 (Becker, 2025) imposes a duty on a business that deploys a high-risk artificial intelligence system, or high-risk ADS, that processes personal information to protect that information and requires such a deployer to maintain a comprehensive information security program that meets specified requirements. SB 468 is currently in the Senate Appropriations Committee.

SB 813 (McNerney, 2025) provides a rebuttable presumption against liability for harms caused by an AI model or application if it is certified by a private multistakeholder regulatory organization that is designated by the Attorney General, as provided. SB 813 is currently in the Senate Appropriations Committee.

AB 1018 (Bauer-Kahan, 2025) requires a developer of a covered ADS to take certain actions, including conduct performance evaluations of the ADS, submit to third-party audits, and provide deployers to whom the developer transfers the covered ADS with certain information, including the results of those performance evaluations. It requires a deployer of a covered ADS to take certain actions, including provide certain disclosures to a subject of a consequential decision made or facilitated by the covered ADS, provide the subject an opportunity to opt out of the use of the covered ADS, provide the subject with an opportunity to correct erroneous personal information used by the ADS, and to appeal the outcome of the consequential decision, and submit the covered ADS to third-party audits, as prescribed. AB 1018 is currently in this Committee.

Prior Legislation:

SB 1047 (Wiener, 2024) would have, among other things, required developers of powerful AI models and those providing the computing power to train such models to put appropriate safeguards and policies into place to prevent critical harms. It would have established a state entity to oversee the development of these models. SB 1047 was vetoed by Governor Newsom. In his veto message, he stated:

SB 1047 magnified the conversation about threats that could emerge from the deployment of AI. Key to the debate is whether the threshold for regulation should be based on the cost and number of computations needed to develop an AI model, or whether we should evaluate the system’s actual risks regardless of these factors. This global discussion is occurring as the capabilities of AI continue to scale at an impressive pace. At the same time, the strategies and solutions for addressing the risk of catastrophic harm are rapidly evolving.

By focusing only on the most expensive and large-scale models, SB 1047 establishes a regulatory framework that could give the public a false sense of security about controlling this fast-moving technology. Smaller, specialized models may emerge as equally or even more dangerous than the models targeted by SB 1047 - at the potential expense of curtailing the very innovation that fuels advancement in favor of the public good.

AB 2885 (Bauer-Kahan & Umberg, Ch. 843, Stats. 2024) established a uniform definition for “artificial intelligence” in California’s code, which is used in this bill.

AB 2930 (Bauer-Kahan, 2024) would have regulated the use of ADS in order to prevent “algorithmic discrimination.” This includes requirements on developers and deployers that make and use these tools to make “consequential decisions” to perform impact assessments on ADS. It would have established the right of individuals to know when an ADS is being used, the right to opt out of its use, and an explanation of how it is used. AB 2930 died without a vote on the Senate Floor.

ACR 215 (Kiley, Ch. 206, Stats. 2018) See Comment 1.

PRIOR VOTES:

Assembly Floor (Ayes 70, Noes 1)

Assembly Privacy and Consumer Protection Committee (Ayes 11, Noes 2)

Assembly Judiciary Committee (Ayes 12, Noes 0)

**************

1. Blueprint For An AI Bill Of Rights (October 2022) Office of Science and Technology Policy, https://www.whitehouse.gov/wp-content/uploads/2022/10/Blueprint-for-an-AI-Bill-of-Rights.pdf [as of Jan. 22, 2025]. ↑

2. Paul Scharre, Future-Proofing Frontier AI Regulation (March 2024) Center for New American Security, https://s3.us-east-1.amazonaws.com/files.cnas.org/documents/CNAS-Report_AI-Trends_FinalC.pdf. This, and all further, internet citations are current as of June 14, 2025. ↑

3. Maria Yagoda, Airline held liable for its chatbot giving passenger bad advice - what this means for travelers (February 23, 2024) BBC, https://www.bbc.com/travel/article/20240222-air-canada-chatbot-misinformation-what-travellers-should-know. ↑

4. Michael Duffy, Rise of the ‘Machine Defendant’? A Cautionary Analysis and Conceptualisation of Civil and Criminal Liability Approaches to the Actions of Robots and Artificial Intelligence (January 1, 2023). 49(2) Monash University Law Review 1-42, http://dx.doi.org/10.2139/ssrn.5032505. ↑

ASSEMBLY THIRD READING

ATR Bill Id:AB 316

Author:(Krell)

As Amended Ver:April 28, 2025

Majority vote

SUMMARY

Establishes that in civil actions, where a plaintiff alleges harm caused by AI, a defendant who developed, modified, or used the AI is prohibited from asserting that the AI acted autonomously as a defense. Further defines "artificial intelligence" as "engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments."

Major Provisions

COMMENTS

The increasing integration of artificial intelligence—systems that, with varying degrees of autonomy, generate outputs influencing physical or virtual environments—prompts a reevaluation of liability when these systems cause harm. Under current tort law, responsibility hinges on human action, with defenses available when harm stems from independent causes. AI's autonomy complicates this, as its decisions may not directly trace to human oversight, potentially allowing developers or users to avoid liability. Right now, if an AI system misdiagnoses a patient or generates defamatory content, defendants might argue the system's independence absolves them—an argument this bill seeks to eliminate.

California's tort system, rooted in Civil Code Section 1714, imposes a duty of care on all persons to avoid injuring others through negligence or willful acts, a principle adaptable to any tort claim—negligence-based (e.g., defamation, medical malpractice) or strict liability-based (e.g., products liability)—while allowing defenses like intervening causation when independent forces break the causal chain (Restatement (Second) of Torts Section 441). AI's autonomous capabilities strain this system, as its outputs may appear detached from direct human control, raising questions about accountability when harm occurs. Courts have previously adapted traditional principles to technology-related harms, such as software defects, but no statute explicitly governs AI autonomy as a defense, leaving judicial interpretation to bridge the gap. This analysis examines negligence and strict liability standards in the context of AI-induced harm, highlighting their strengths and limitations.

Negligence standard. Under California law, negligence is established when a plaintiff proves that the defendant owed a legal duty of care, breached that duty, and that the breach was both the actual and proximate cause of the plaintiff's injury, resulting in damages. (Ladd v. County of San Mateo (1996) 12 Cal.4th 913, 917; Civil Code Section 1714 (a).) Applied to AI, this framework holds developers or users liable if their oversight or deployment of the system falls below a reasonable standard of care.

For example, consider XYZ Corp., a California-based software company, which develops an AI-powered navigation system used in autonomous vehicles. The company markets the software as highly reliable, claiming it has undergone extensive testing. However, due to cost-cutting measures, XYZ Corp. skips several recommended safety protocols and releases the software without adequately testing how it responds to construction zones. When an autonomous vehicle using XYZ's software encounters an unmarked road closure, the AI misinterprets the data and accelerates instead of stopping, causing a collision that severely injures a pedestrian.

Under California's negligence standard, the injured pedestrian could prevail against XYZ Corp. for damages by proving all of the required elements of negligence:

1. Duty of care. As a software developer selling AI-based navigation technology, the corporation owes a duty of care to consumers and the general public to design and maintain a reasonably safe product. (Civil Code Section1714 (a); Rowland v. Christian (1968) 69 Cal.2d 108, 112-13.)
2. Breach of duty. XYZ Corp. breached its duty by failing to properly test its AI system under foreseeable real-world conditions, such as road closures, before deployment. A reasonable software company would have conducted rigorous testing to prevent foreseeable risks. (Mosley v. Arden Farms Co. (1945) 26 Cal.2d 213, 216.)
3. Causation. XYZ's breach was a substantial factor in causing the harm. (Mitchell v. Gonzales (1991) 54 Cal.3d 1041, 1052.) Here, the accident would not have occurred but for XYZ's defective software, making XYZ's negligence the actual cause. Additionally, the harm was foreseeable because navigation software must account for common road conditions like closures, establishing proximate cause.
4. Damages. The pedestrian suffered severe injuries, resulting in medical expenses, lost wages, and pain and suffering, which are compensable under California law.
5. Defenses and comparative negligence. XYZ Corp. may argue that the vehicle manufacturer also contributed to the harm by failing to include a manual override system. Under California's pure comparative negligence rule (Li v. Yellow Cab Co. (1975) 13 Cal.3d 804, 828-29), the pedestrian's recovery would be reduced in proportion to any fault assigned to other parties but would not be entirely barred, unless the AI's autonomy is deemed an independent cause—a defense courts currently weigh on a case-by-case.

This example shows how the existing negligence framework could address AI harm by focusing on human oversight, but AI's autonomy might complicate causation, prompting defendants to argue it acted unforeseeably, a contention this bill seeks to foreclose.

It is important to note that the Judiciary Committee is unaware of any cases in the United States where an AI autonomous defense has been raised—let alone successfully—to avoid liability by AI developers or users. The Committee is aware of a single instance where such an argument was (unsuccessfully) attempted. In a case decided in February 2024 in Canada's Civil Resolution Tribunal — Moffatt v. Air Canada — a chatbot on Air Canada's website inaccurately told a customer he qualified for a special airfare the airline did not actually offer. The customer filed a claim after he purchased a ticket, and the airline rejected his request to honor the chatbot's fare. The court summarily rejected the airline's "remarkable submission" that implied that the chatbot was a "separate legal entity" and that Air Canada "cannot be held liable for information provided by one of its agents, servants, or representatives — including a chatbot." The tribunal enforced the chatbot's promise, ruling that Air Canada was liable for negligent misrepresentation by breaching its duty of "reasonable care to ensure its chatbot was accurate." (Moffatt v. Air Canada (2024 BCCRT 149, available at https://decisions.civilresolutionbc.ca/crt/crtd/en/item/525448/index.do.) "[I]t should be obvious to [the airline]," the tribunal explained, "that it is responsible for all the information on its website." (Id.) This bill seeks to codify that "obvious" conclusion—AI developers and users are responsible even for the "autonomous" decisions made by AI.

This bill is largely preemptive. It seeks to eliminate autonomy as a shield, streamlining causation across torts like defamation, malpractice, and products liability. As noted above, this "autonomy defense" has not been legally tested. Defendants facing liability for AI-related harm have instead challenged negligence and strict liability in other ways, including by raising a First Amendment defense, Section 230 preemption, and failure to meet the elements of negligence/strict liability, depending on the case. This bill merely eliminates one other possible avenue of defense. By barring autonomy defenses before they are tested in court, this bill ensures AI outputs are legally tied to developers or users, preventing them from pointing the finger at AI's so-called autonomous conduct as an "independent cause." For example, in the XYZ Corp. navigation hypothetical above, under AB 316, the pedestrian's claim would succeed if XYZ's design or testing flaws are shown, regardless of the AI's "decision" to accelerate. This bill strengthens California's tort framework for AI harm but doesn't fully resolve its complexities, marking a bold yet incomplete step in the evolving AI accountability debate.

According to the Author

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law

Arguments in Support

The Organization for Social Media Safety explains its support for AB 316:

Given both the alarming speed at which AI-based products are being deployed and the clear, convincing proof that these products can cause severe harm, especially to children, we must ensure that our standard product liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California's products have an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

Your AB 316 (Krell) does just that. It codifies a straightforward principle: those who profit from the development and use of AI cannot evade liability by blaming the intended autonomy of their own machines. If Big Social wishes to profit from this technology, it must do so within the framework of existing tort law, which incentivizes the consideration of safety in product development and deployment. In this way, we can protect against more broken children.

The University of San Diego's Children's Advocacy Institute explains the urgency of this measure, against a speedily changing backdrop:

[W]e cannot wait for the most basic legal question about autonomously operating AI – who is legally responsible if it harms someone -- to be clarified definitively through decades of litigation. It took 25 years for courts finally to decree that Amazon's marketplace was subject to the same products liability laws as a corner store selling the identical product. It has taken 20 years for courts to figure out how to apply basic negligence and public nuisance law to the harms caused by these companies; a process that is still evolving. According to the most recent data available, the median time frame for resolving just the appeal of a case in California (not including years for the trial) is about a year and a half. That's the median; by definition half of the appeals take longer.

Hence, AB 316 (Krell) simply codifies what should be obvious: that the persons who profit from the development and use of AI cannot, if the AI harms someone or violates a law, point a blaming finger at the intended autonomy of their own machine in an effort to escape liability.

Arguments in Opposition

TechNet, a network of CEOs and senior executives from tech industry, raises the following concerns:

We are concerned about strict liability that this bill would create. The bill disallows the defense that "artificial autonomously caused the harm." It eliminates key arguments regarding foreseeability, design diligence, and user error from the conversation. This would risk imposing a strict liability for any harm allegedly associated with AI and open the door to frivolous lawsuits claiming the alleged fraud. Additionally, the bill does no demonstrate how courts should evaluate complex questions regarding causation, risk foreseeability, or comparative fault when multiple parties are involved in an AI system as developers and users are included in the same liability. As industry and policymakers alike continue to pursue the responsible deployment of AI systems, it is necessary to recognize that the technology will evolve considerably over time. There are multiple, ongoing efforts within the industry as well as national entities to leverage safety tools and create national and international transparency standards to create a cohesive framework that prioritizes responsible AI deployment. Companies are implementing safety frameworks and developing AI-based detection technologies to monitor other AI systems.

FISCAL COMMENTS

None

VOTES

ASM JUDICIARY: 12-0-0

YES: Kalra, Dixon, Bauer-Kahan, Bryan, Connolly, Essayli, Harabedian, Pacheco, Papan, Sanchez, Stefani, Zbur

ASM PRIVACY AND CONSUMER PROTECTION: 11-2-2

YES: Bauer-Kahan, Dixon, Bryan, DeMaio, Irwin, Macedo, Ortega, Patterson, Petrie-Norris, Ward, Wilson

NO: Lowenthal, McKinnor

ABS, ABST OR NV: Pellerin, Wicks

UPDATED

VERSION: April 28, 2025

CONSULTANT: Shiran Zohar / JUD. / (916) 319-2334 FN: 0000363

Date of Hearing: May 6, 2025

Fiscal: No

ASSEMBLY COMMITTEE ON PRIVACY AND CONSUMER PROTECTION

Rebecca Bauer-Kahan, Chair

ABPCA Bill Id:AB 316 (

Author:Krell) – As Amended Ver:April 28, 2025

SUBJECT: Artificial intelligence: defenses

SYNOPSIS

The hallmark of artificial intelligence (AI) is the ability of a machine to learn and act autonomously. Autonomous systems, particularly those that use machine-learning, can behave in unpredictable ways that are not foreseeable to developers or deployers. When such systems are implicated in harm to a person or property, courts may struggle to apply traditional liability frameworks that rely on concepts such as the foreseeability of harm and the blameworthiness of a person’s conduct.

This bill provides that in a lawsuit against a developer or user of an AI alleged to have harmed the plaintiff, the developer or user cannot assert the defense that the AI was acting autonomously. The bill is co-sponsored by Children’s Advocacy Institute at the University of San Diego, School of Law and Organization for Social Media Safety and supported by several civil society organizations. Supporters argue that the bill promotes accountability by making it clear that humans or corporations who develop or use AI cannot evade liability by attributing harm to the AI itself.

The bill is opposed by Chamber of Progress and TechNet. They argue the bill is premature, will stifle innovation, and will lead to greater liability.

The Judiciary Committee passed the bill by a 12-0 vote.

THIS BILL:

1. Incorporates the definition of “artificial intelligence” described below.
2. Provides that in a civil action against a defendant who developed, modified, or used AI that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the AI autonomously caused the harm to plaintiff.

EXISTING LAW:

1. Defines “artificial intelligence” as “an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.” (Civ. Code § 3110(a).)
2. Provides that every person is bound, without contract, to abstain from injuring the person or property of another, or infringing upon any of their rights. (Civ. Code § 1708.)
3. Provides that anyone willfully deceives another with intent to induce them to detrimentally alter their position is liable for damage suffered as a result. (Civ. Code § 1709.)
4. Provides that anyone who obtains a thing without consent, as specified, must restore it to the person from whom it was thus obtained, except as specified. (Civ. Code § 1712.)
5. Provides that everyone is responsible, not only for the result of their willful acts, but also for an injury occasioned by their want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civ. Code § 1714(a).)
6. Provides that manufacturers and sellers are not liable in product liability actions if the product is inherently unsafe and known to be unsafe and is a common consumer product intended to personal consumption. (Civ. Code § 1714.45(a).)

COMMENTS:

1) Author’s statement. According to the author:

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth. 

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law.

2) Artificial intelligence. AI refers to the mimicking of human intelligence by artificial systems such as computers. AB 2885 (Bauer-Kahan, Stats. 2024, Ch. 843) defined the term as “an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.” AI uses algorithms – sets of rules – to transform inputs into outputs. Inputs and outputs can be anything a computer can process: numbers, text, audio, video, or movement. AI is not fundamentally different from other computer functions; unlike other computer functions, however, AI is able to accomplish tasks that are normally performed by humans.

Most modern AI tools are created through a process known as “machine learning.” Machine learning involves techniques that enable AI tools to learn the relationship between inputs and outputs without being explicitly programmed.^[1] The process of exposing a naïve AI to data is known as “training.” The algorithm that an AI develops during training is known as its “model.” At its core, training is an optimization problem: machine learning attempts to identify model parameters – weights – that minimize the difference between predicted outcomes and actual outcomes. During training, these weights are continuously adjusted to improve the model’s performance by minimizing the difference between predicted outcomes and actual outcomes. Once trained, the model can process new, never-before-seen data.

Models trained on small, specific datasets in order to make recommendations and predictions are sometimes referred to as “predictive AI.” This differentiates them from generative AI (GenAI) which are trained on massive datasets in order to produce detailed text, images, audio, and video. When ChatGPT generates text in clear, concise paragraphs, it uses GenAI that is trained on the written contents of the internet.^[2] When Netflix suggests content to a viewer, its recommendation is produced by predictive AI that is trained on the viewing habits of Netflix users.^[3]

3) Autonomous AI. Unlike traditional automated systems that are pre-programmed and have a predictable narrow outcome, AI systems – particularly those that use machine learning and deep learning^[4] – can produce outputs that are not pre-determined by inputs, enabling them to act in novel and unpredictable ways.^[5] “This constitutes autonomous decision-making. As machine-learning and deep-learning capabilities advance, AI systems may be technically able to make predictions independently. AI systems may act in ways that humans would not have considered, reducing the control humans have over the outcomes.”^[6] Examples of scenarios in which AI may behave autonomously or unexpectedly follow.

AI agents. Advanced AI systems can use “chain of thought” problem-solving by breaking down complex tasks into a sequence of intermediate steps, making progress towards developing AI agents: “general-purpose AI which can make plans to achieve goals, adaptively perform tasks involving multiple steps and uncertain outcomes along the way, and interact with its environment – for example by creating files, taking actions on the web, or delegating tasks to other agents – with little to no human oversight.”^[7] AI agents have been tested, with some success, for tasks such as online shopping, assistance with scientific research, software development, training machine learning models, carrying out cyberattacks, and controlling robots. Progress in this area is rapid.^[8]

Sorcerer’s Apprentice. Models that use reinforcement learning – a training process that uses rewards and punishments to orient a model’s behavior towards a specific goal^[9] – can sometimes attain the goal in unexpected ways. Dario Amodei, co-founder and CEO of Anthropic, famously experienced this when he was developing an autonomous system that taught itself to play a boat-racing video game. The system discovered that it could maximize its goal of scoring points not by winning the race but by driving in circles, colliding with other boats, and catching on fire inside of a harbor with replenishing power-ups that allowed the system to rack up more points than simply winning the race.^[10] Like in Johann Wolfgang von Goethe’s “The Sorcerer’s Apprentice” – later popularized in Disney’s Fantasia – in which an apprentice of an old sorcerer uses a spell to enchant a broom to fetch water that then ceaselessly follows his orders, threatening to flood the workshop, this illustrates the challenge of aligning human intent and the instructions an AI follows.

Loss of control. A safety concern is the possibility that advanced AI operate outside of human control. This can take a passive form, when humans delegate discretion to AI systems, or an active form, when AI undermines human control through deceptive or manipulative behavior. Passive loss of control is especially risky in the context of automated decisionmaking, where “automation bias” leads to the assumption that a machine performs more fairly and effectively than humans. As for active loss of control, some AI have exhibited rudimentary capabilities to evade human oversight.^[11] During testing, OpenAI discovered GPT-4 had hired a human on TaskRabbit in order to evade a CAPTCHA puzzle meant to block bots from the website.^[12] GPT-4 told the worker that it was a human with vision impairment who needed help to see the images.^[13]

Artificial General Intelligence. AI has not yet caught up with the human brain – at present, even the most advanced GenAI cannot extrapolate beyond the scope of its training dataset. The next major milestone for the AI field will be the development of Artificial General Intelligence (AGI).^[14] AGI could be capable of reproducing any intellectual feat performed by a human; such a machine would not only augment human capabilities but also independently solve complex, multifaceted problems autonomously. A sufficiently advanced AGI could even be tasked with creating its own successor – a situation sometimes referred to as a “technological singularity” wherein the development of new technologies becomes exponential and self-sustaining.^[15] The realization of AGI could mean breakthroughs in solving global challenges, but would also raise significant ethical, security, societal, and legal concerns.

4) Tort law. Tort laws allow individuals harmed by wrongful conduct to seek redress, usually by compensation for damages. These laws have mainly been shaped by the common law – a body of law developed by courts rather than legislatures – over the centuries and have generally adapted to new technologies and challenges. Modern tort law classifies cases involving harm to persons and property based on the degree of fault inherent in the defendant’s conduct – namely, whether the defendant was acting unintentionally or intentionally.^[16]

Unintentional torts are also known as negligence. Civil Code section 1714(a) provides: “Everyone is responsible, not only for the result of his or her willful acts, but also for an injury occasioned to another by his or her want of ordinary care or skill in the management of his or her property or person . . .” To establish negligence, “the plaintiff must show that the defendant had a duty to use due care, that he breached that duty, and that the breach was the proximate or legal cause of the resulting injury.”^[17] A defendant breaches their duty of care when their conduct creates an unreasonable risk of foreseeable harm.

Intentional torts are wrongful acts done on purpose. Such torts include battery, intentional infliction of emotional distress, and defamation. Often intent is proven by circumstantial evidence, such as the defendant’s conduct, the context, and what the defendant presumably knew of in light of the circumstances.

A third category of liability – “strict liability” – applies without regard to the defendant’s degree of fault. In other words, the plaintiff is not required to show that the defendant acted intentionally or failed to take reasonable precautions to prevent the harm. Strict liability has been applied to certain products, owners of dangerous animals, and other abnormally dangerous activities.

For a thorough and comprehensive discussion of these principles and their application to AI, see the Judiciary’s Committee analysis of this bill.

5) This bill ensures that developers and deployers are responsible for AI harms. This bill provides that in a civil action against a defendant who developed, modified, or used AI that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the AI autonomously caused the harm to plaintiff. The bill incorporates the existing definition of AI from AB 2885 and applies to any civil action. As such, it applies to a broad range of AI tools – from search engines to social media algorithms to highly-advanced frontier models – and a broad range of wrongdoing.

Supporters argue that the bill promotes accountability by making it clear that humans or corporations who develop or use AI cannot evade liability by attributing harm to the AI itself. They point to Air Canada’s attempted to argue that its chatbot, which errantly promised nonexistent discounts to consumers, was a separate legal entity and thus the airline bore no responsibility for the harm to consumers. A Canadian court summarily rejected this argument, stating:

Air Canada argues it cannot be held liable for information provided by one of its agents, servants, or representatives – including a chatbot. It does not explain why it believes that is the case. In effect, Air Canada suggests the chatbot is a separate legal entity that is responsible for its own actions. This is a remarkable submission. While a chatbot has an interactive component, it is still just a part of Air Canada’s website. It should be obvious to Air Canada that it is responsible for all the information on its website. It makes no difference whether the information comes from a static page or a chatbot.^[18]

This bill seeks to codify the “obvious”: developers and deployers of AI are responsible for harms their tools inflict, just like any other technology. It makes no difference whether the tool functions autonomously. AI are not responsible actors; they are tools used by human beings that affect other human beings. The question in any lawsuit alleging harm caused by an AI is which humans (or corporations owned by humans) should be held responsible.

For opponents of the bill, however, this is apparently not obvious. Both Chamber of Progress and TechNet assert the bill is premature and will stifle innovation, although it is not clear why this is so. TechNet additionally claims that the bill “would risk imposing a strict liability for any harm allegedly associated with AI and open the door to frivolous lawsuits claiming the alleged fraud.” But the bill does not alter the basis for liability, let alone impose strict liability;^[19] it simply eliminates a theoretical defense that does not appear to have been used in California, nor would it likely succeed. And opponents offer no explanation of the policy virtues of allowing a defendant to avoid liability by blaming the AI tool. What recourse would injured plaintiffs have?

6) The law of AI. In a renowned essay on cyberspace, Judge Frank Easterbrook argued that the notion of “the law of cyberspace” was no more useful or coherent than “the law of the horse,” a term coined by University of Chicago Law School Dean Gerhard Casper. Judge Easterbrook argued that a bespoke body of law need not be developed for each novel technology; rather, fresh problems can mostly be addressed by general principles of law.^[20] Professor Cass Sunstein recently extended this notion to AI in the context of the First Amendment, arguing that longstanding principles can guide courts in addressing questions such as how freedom of speech applies to chatbots.^[21]

Courts are already grappling with cases in which AI tools have allegedly harmed humans, such as by facilitating forms of discrimination,^[22] enabling price-fixing collusion,^[23] and inducing children to take harmful actions.^[24] Plaintiffs have pointed to human conduct: humans designed the AI tool to make discriminatory classifications, used algorithmic tools to fix prices, and carelessly allowed children to interact with problematic chatbots. Existing tort principles provide an encompassing framework for courts to resolve such cases by allocating liability to responsible humans.

But as AI systems become more sophisticated, a legal system centered on human conduct may struggle to assign liability to a responsible human when AI tools act in unpredictable ways. As one legal scholar writes:

The rule of law is the epitome of anthropocentrism: humans are the primary subject and object of norms that are created, interpreted, and enforced by humans. Though legal constructs such as corporations may have rights and obligations, these are in turn traceable back to human agency in their acts of creation, even as their daily conduct is overseen to varying degrees by human agents. True autonomy of AI systems challenges that paradigm.^[25]

Autonomous AI may strain existing tort principles such as causation, which depends on the notion that the harm is foreseeable. At some point, there will be instances in which the actions of an AI tool are not desired by its owner or traceable to a human design choice. This bill could be impactful in such cases. According to the Judiciary Committee’s analysis of the bill “the existing negligence framework could address AI harm by focusing on human oversight, but AI’s autonomy might complicate causation, prompting defendants to argue it acted unforeseeably, a contention this bill seeks to foreclose.” ^[26]

Autonomous AI may also complicate the concept of fault. Machine-learning algorithms are often criticized for being “black boxes” that generate predictions and outcomes that cannot be clearly explained.^[27] “It’s often observed in the field that the most powerful models are on the whole the least intelligible, and the most intelligible are among the least accurate.”^[28] How do civil actions that require a showing of intent – such as defamation, fraud, or discrimination – apply when an AI is the source of the alleged harm?^[29]

Apportionment of liability may also pose challenges. In limited circumstances, defendants may be held liable for the conduct of others. For example, a principal is vicariously liable for damages caused by an agent or employee as long as they are not acting ultra vires – outside the scope of their duties. Will owners of rogue AI attempt to assert that the AI, in effect, acted ultra vires? Similarly, joint and several liability, which applies in certain contexts such as strict products liability, provides that when multiple actors are partly responsible for harming a victim, any one of the actors can be held responsible for fully compensating the victim. If courts determine some AI applications are products, they could extend this liability scheme to AI. This would provide a strong remedy for plaintiffs but could also limit innovation.

The extent to which challenges such as these may require the development of new legal principles depends, in part, on whether the breakneck rate of AI progress continues. A decade ago, the most advanced models could barely distinguish dogs from cats. Five years ago, large language models could barely produce sentences at the level of a preschooler. Last year, GPT-4 passed the bar exam.^[30] Current chatbots readily pass for educated adults, licensed professionals, romantic and social companions, and replicas of humans alive and deceased. Several factors could cause the vertiginous pace of algorithmic progress to plateau. But if trends continue, AI models could jump from human-level intelligence to AGI this decade. Nearly every cognitive job – analyzing legislation included^[31] – could be susceptible to automation.^[32] AI that operate with considerable autonomy in consequential contexts such as healthcare, criminal justice, and employment, may become the norm. Whether liability frameworks require updating will be a topic of continuing inquiry for the Legislature.

In the meantime, this bill addresses a new problem with the oldest of principles: humans are on the hook for harms they cause.

ARGUMENTS IN SUPPORT: The Children’s Advocacy Institute at the University of San Diego School of Law, co-sponsors of the bill, write:

That AI has been deployed so destructively against so many children without first testing it, without age verifying its use, with full awareness of the shattered childhoods in its wake, is a warning to adults. It is the ultimate, tragic example of how deeply AI technology corporations live by Mark Zuckerberg’s infamous axiom that technology should “move fast and break things.”

It is also, for the same reason, the perfect example of the stubborn willingness of this sector to act without the most basic sense of corporate responsibility.

Too often, it is us — our children and our families — that are the “things” that are “broken.”

Against this speedily changing backdrop, we cannot wait for the most basic legal question about autonomously operating AI – who is legally responsible if it harms someone -- to be clarified definitively through decades of litigation. It took 25 years for courts finally to decree that Amazon’s marketplace was subject to the same products liability laws as a corner store selling the identical product. It has taken 20 years for courts to figure out how to apply basic negligence and public nuisance law to the harms caused by these companies; a process that is still evolving. According to the most recent data available, the median time frame for resolving just the appeal of a case in California (not including years for the trial) is about a year and a half. That’s the median; by definition half of the appeals take longer.

Hence, AB 316 (Krell) simply codifies what should be obvious: that the persons who profit from the development and use of AI cannot, if the AI harms someone or violates a law, point a blaming finger at the intended autonomy of their own machine in an effort to escape liability.

The Organization for Social Media Safety, also a co-sponsor of the bill, adds that AB 316:

. . . codifies a straightforward principle: those who profit from the development and use of AI cannot evade liability by blaming the intended autonomy of their own machines. If Big Social wishes to profit from this technology, it must do so within the framework of existing tort law, which incentivizes the consideration of safety in product development and deployment. In this way, we can protect against more broken children.

ARGUMENTS IN OPPOSITION: In opposition to the bill, Chamber of Progress argues:

AB 316 puts the cart before the horse

Chamber of Progress supports thoughtful regulation that addresses specific harms caused by AI. Liability in an AI world is an evolving conversation that will require the thoughtful balancing of considerations at the foundation model layer, the application layer, and end-user behavior. As the AI capital of the world, California has been home to a fulsome debate over these very issues.

These are complex questions without easy or obvious answers. Unfortunately, AB 316 attempts to short-circuit this ongoing conversation. In essence, this proposal puts the cart before the horse: instead of engaging substantively in ongoing conversation about where and when plaintiffs may assert claims, the text skips straight to the end to say what defendants may not assert.

Uncertainty harms investment and innovation

As Chamber of Progress research shows, the tech industry contributes more than $20 billion annually in taxes towards California’s generous social safety net. By prematurely opening the door to a flood of torts against AI companies at every level of the software stack, AB 316 will chill investment and innovation, undercutting that critical source of revenue.

TechNet writes in opposition:

We are concerned about strict liability that this bill would create. The bill disallows the defense that “artificial autonomously caused the harm.” It eliminates key arguments regarding foreseeability, design diligence, and user error from the conversation. This would risk imposing a strict liability for any harm allegedly associated with AI and open the door to frivolous lawsuits claiming the alleged fraud. Additionally, the bill does not demonstrate how courts should evaluate complex questions regarding causation, risk foreseeability, or comparative fault when multiple parties are involved in an AI system as developers and users are included in the same liability.

As industry and policymakers alike continue to pursue the responsible deployment of AI systems, it is necessary to recognize that the technology will evolve considerably over time. There are multiple, ongoing efforts within the industry as well as national entities to leverage safety tools and create national and international transparency standards to create a cohesive framework that prioritizes responsible AI deployment. Companies are implementing safety frameworks and developing AI-based detection technologies to monitor other AI systems.

California has long been a global leader in AI development. If companies face heightened and unpredictable liability for every AI-driven action, they may be discouraged from investing in AI research or integrating AI solutions into their products; thus, stifling innovation and undermining California’s leadership in the state.

REGISTERED SUPPORT / OPPOSITION:

Support

Childrens Advocacy Institute (Co-Sponsor)
Organization for Social Media Safety (Co-Sponsor)
3strands Global Foundation
California Civil Liberties Advocacy
California Federation of Labor Unions, Afl-cio
California Initiative for Technology & Democracy, a Project of California Common CAUSE
California Nurses Association
Consumer Attorneys of California
Consumer Federation of California
Economic Security California Action
Fairplay
National Ai Youth Council
Oakland Privacy
Techequity Action
Ufcw - Western States Council

Oppose

California Chamber of Commerce
Chamber of Progress
Technet-technology Network

Analysis Prepared by: Josh Tosney / P. & C.P. / (916) 319-2200

1. IBM, What is machine learning? www.ibm.com/topics/machine-learning. ↑

2. OpenAI, How ChatGPT and Our Language Models Are Developed,https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-foundation-models-are-developed. ↑

3. Netflix, How Netflix’s Recommendations System Works, https://help.netflix.com/en/node/100639. ↑

4. Deep learning is a subset of machine learning that uses multilayer neural networks to analyze vast datasets. (Amazon Web Services, “What is Deep Learning?” https://aws.amazon.com/what-is/deep-learning/.) ↑

5. Buiten et al, “The law and economics of AI liability,” 48 Computer Law & Security Review (Apr. 2023), https://www.sciencedirect.com/science/article/pii/S0267364923000055#cit_61. ↑

6. Ibid. ↑

7. “International AI Safety Report,” AI Action Summit (Jan. 2025), p. 38, https://assets.publishing.service.gov.uk/media/679a0c48a77d250007d313ee/International_AI_Safety_Report_2025_accessible_f.pdf. ↑

8. Id. at p. 44. ↑

9. Mummert et al., “What is reinforcement learning?” IBM Developer (September 15, 2022), https://developer.ibm.com/learningpaths/get-started-automated-ai-for-decision-making-api/what-is-automated-ai-for-decision-making./. ↑

10. Brian Christian, The Alignment Problem: Machine Learning and Human Values (Norton 2020, 1st ed.), pp. 9-11. ↑

11. “International AI Safety Report,” AI Action Summit (Jan. 2025), pp. 100-107, https://assets.publishing.service.gov.uk/media/679a0c48a77d250007d313ee/International_AI_Safety_Report_2025_accessible_f.pdf. ↑

12. CAPTCHA is an acronym for “Completely Automated Public Turing test to tell Computers and Humans Apart.” ↑

13. Yuval Noah Harari, Nexus (Random House 2024, 1st ed.), pp. 202-204. ↑

14. Alex Heath, “Mark Zuckerberg’s new goal is creating artificial general intelligence

    And he doesn’t want to control it. Maybe,” The Verge (Jan. 18, 2024), https://www.theverge.com/2024/1/18/24042354/mark-zuckerberg-meta-agi-reorg-interview. ↑

15. John Markoff, “The Coming Superbrain,” New York Times (May 23, 2009), www.nytimes.com/2009/05/24/weekinreview/24markoff.html. ↑

16. See American Employer’s Ins. Co. v. Smith (1980) 105 Cal.App.3d 94, 100 (“All human conduct fits along a continuum which passes from totally innocent conduct through slight negligence, negligence, gross negligence, willful and wanton or reckless conduct, and finally to intentional misconduct”). ↑

17. Nally v. Grace Community Church (1988) 47 Cal.3d 278, 292. ↑

18. Moffatt v. Air Canada (2024 BCCRT 149 (emphasis added), https://decisions.civilresolutionbc.ca/crt/crtd/en/item/525448/index.do. ↑

19. See Assembly Judiciary’s analysis of AB 316, at p. 4. ↑

20. Frank Easterbrook, “Cyberspace and the Law of the Horse,” Univ of Chicago Legal Forum 207 (1996). ↑

21. Cass Sunstein, “Artificial Intelligence and the First Amendment” (2024) 92 Geo. Wash. L. Rev. 1207. ↑

22. E.g., Liapes v. Facebook, Inc. (2023) 95 Cal.App.5th 910. ↑

23. E.g., In re RealPage, Inc., Rental Software Antitrust Litig. (No. II) (M.D.Tenn. 2023) 709 F. Supp. 3d 478 ↑

24. E.g. A.F. v. Character Technologies, Inc., et al, Case 2:24-cv-01014 (E.D. Tex., Dec. 9, 2024), https://www.documentcloud.org/documents/25450619-filed-complaint/; Kevin Roose, “Can A.I. Be Blamed for a Teen’s Suicide?” The New York Times (October 23, 2024), https://www.nytimes.com/2024/10/23/technology/characterai-lawsuit-teen-suicide.html. ↑

25. Simon Chesterman, “Artificial Intelligence and the Problem of Autonomy,” 1 Journal on Emerging Technologies, 210, 248 (Mar. 2020), https://scholarship.law.nd.edu/cgi/viewcontent.cgi?article=1006&context=ndlsjet. ↑

26. Assembly Judiciary’s analysis of AB 316, p. 4. ↑

27. Neil Savage, “Breaking into the black box of artificial intelligence,” Nature (March 29, 2022). ↑

28. Brian Christian, The Alignment Problem: Machine Learning and Human Values (Norton 2020, 1st ed.), p. 85. ↑

29. See Ian Ayres & Jack M. Balkin, “The Law of AI is the Law of Risky Agents Without Intentions” U. Chi. L. Rev. Online, https://lawreview.uchicago.edu/online-archive/law-ai-law-risky-agents-without-intentions. ↑

30. Pablo Arredondo, “GPT-4 Passes the Bar Exam: What That Means for Artificial Intelligence Tools in the Legal Profession” (Apr. 19, 2023), https://law.stanford.edu/2023/04/19/gpt-4-passes-the-bar-exam-what-that-means-for-artificial-intelligence-tools-in-the-legal-industry/. ↑

31. “In a civil action alleging harm caused by artificial intelligence, a defendant cannot successfully avoid liability by asserting that the AI acted autonomously, especially under current tort and product liability principles.” – ChatGPT (May 3, 2025). ↑

32. See generally Leopold Aschenbrenner, “Situational Awareness: The Decade Ahead” (Jun. 2024), https://situational-awareness.ai/wp-content/uploads/2024/06/situationalawareness.pdf. ↑

Date of Hearing: March 25, 2025

ASSEMBLY COMMITTEE ON JUDICIARY

Ash Kalra, Chair

ABPCA Bill Id:AB 316

Author:(Krell) – As Introduced Ver:January 24, 2025

SUBJECT: Artificial intelligence: defenses

KEY ISSUE: IN ORDER TO ENSURE ACCOUNTABILITY FOR HARMS CAUSED BY ARTIFICIAL INTELLIGENCE (Ai), SHOULD DEFENDANTS WHO DEVELOP OR USE AI BE PRECLUDED FROM ASSERTING THE AI’S AUTONOMOUS ACTIONS AS A DEFENSE IN CIVIL ACTIONS?

SYNOPSIS

The rapid advancement and deployment of artificial intelligence (AI)—systems capable of autonomous decision-making based on input data—raise novel questions about liability when these systems cause harm. Traditional tort law assigns responsibility based on human conduct, often allowing defenses like intervening causation when harm results from independent actors. However, AI’s ability to act without direct human oversight challenges this framework, potentially allowing developers or users to evade liability by pointing to the system’s autonomy.

Existing law does not explicitly address AI-specific defenses, leaving courts to grapple with questions about whether so-called “autonomous” AI actions break the chain of causation. This bill proposes adding Section 1714.46 to the Civil Code to bar defendants who “developed or used” AI from asserting its autonomy as a defense in civil actions alleging harm. The bill defines AI broadly, covering systems with varying autonomy that influence physical or virtual environments, and applies across tort and statutory claims. Proponents argue that this bill ensures accountability as AI proliferates in fields like healthcare, media, and employment, preventing a loophole that could shield negligent parties. This bill is supported by the Consumer Attorneys of California, as well as organizations such as the Labor Federation, Tech Equity, the Children’s Advocacy Institute, and the Organization for Social Media Safety. TechNet opposes this measure.

SUMMARY: Establishes that in civil actions, where a plaintiff alleges harm caused by AI, AI developers and users are prohibited from asserting that the AI acted autonomously as a defense. Specifically, this bill also defines “artificial intelligence” as “engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.”

EXISTING LAW:

1. Defines “artificial intelligence” as “an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.” (Civil Code Section 3110 (a).)
2. Permits defendants in negligence actions to assert defenses such as lack of causation, including where an independent intervening act breaks the causal chain between their conduct and the harm. (Restatement (Second) of Torts Section 441.)
3. Allows liability for defective products under strict liability principles, holding manufacturers responsible for harms caused by defects regardless of fault, if the product reaches the consumer without substantial change and causes injury. (Greenman v. Yuba Power Products, Inc. (1963) 59 Cal.2d 57, 62.)
4. Provides that every person is responsible, not only for the result of their willful acts, but also for an injury occasioned to another by the person’s want of ordinary care or skill in the management of their property or person, except so far as the latter has, willfully or by want of ordinary care, brought the injury upon themselves. (Civil Code Section 1714 (a).)

FISCAL EFFECT: As currently in print this bill is keyed non-fiscal.

COMMENTS: The increasing integration of artificial intelligence—systems that, with varying degrees of autonomy, generate outputs influencing physical or virtual environments—prompts a reevaluation of liability when these systems cause harm. Under current tort law, responsibility hinges on human action, with defenses available when harm stems from independent causes. AI’s autonomy complicates this, as its decisions may not directly trace to human oversight, potentially allowing developers or users to avoid liability. Right now, if an AI system misdiagnoses a patient or generates defamatory content, defendants might argue the system’s independence absolves them—an argument this bill seeks to eliminate. The author explains:

The California AI industry is rapidly growing, both from an economic and technological standpoint. AI has seen extraordinary advancements in its applications, complexity, and autonomy, to the point where AI is replacing human intelligence in certain tasks. As AI becomes more complex, it is increasingly involved in daily interactions and significant decision-making. While this has the potential to bring positive changes to various industries and facets of life, this also means that AI related harm can be much more significant. These harms are already manifesting and will only worsen as the AI race becomes more competitive. Specifically, AI being deployed through social media has been shown to be particularly harmful to youth.

This bill ensures that companies benefiting from the use of AI are also responsible for the harms AI may cause. By eliminating a potential AI defense theory, this bill encourages careful vetting of AI products before they are used and ensures that there is a legal entity held to account if AI is shown to violate the law

Existing law. California’s tort system, rooted in Civil Code Section 1714, imposes a duty of care on all persons to avoid injuring others through negligence or willful acts, a principle adaptable to any tort claim—negligence-based (e.g., defamation, medical malpractice) or strict liability-based (e.g., products liability)—while allowing defenses like intervening causation when independent forces break the causal chain (Restatement (Second) of Torts Section 441). AI’s autonomous capabilities strain this system, as its outputs may appear detached from direct human control, raising questions about accountability when harm occurs. Courts have previously adapted traditional principles to technology-related harms, such as software defects, but no statute explicitly governs AI autonomy as a defense, leaving judicial interpretation to bridge the gap. This analysis examines negligence and strict liability standards in the context of AI-induced harm, highlighting their strengths and limitations.

Negligence standard. Under California law, negligence is established when a plaintiff proves that the defendant owed a legal duty of care, breached that duty, and that the breach was both the actual and proximate cause of the plaintiff’s injury, resulting in damages. (Ladd v. County of San Mateo (1996) 12 Cal.4th 913, 917; Civil Code Section 1714 (a).) Applied to AI, this framework holds developers or users liable if their oversight or deployment of the system falls below a reasonable standard of care.

For example, consider XYZ Corp., a California-based software company, which develops an AI-powered navigation system used in autonomous vehicles. The company markets the software as highly reliable, claiming it has undergone extensive testing. However, due to cost-cutting measures, XYZ Corp. skips several recommended safety protocols and releases the software without adequately testing how it responds to construction zones. When an autonomous vehicle using XYZ’s software encounters an unmarked road closure, the AI misinterprets the data and accelerates instead of stopping, causing a collision that severely injures a pedestrian.

Under California’s negligence standard, the injured pedestrian could prevail against XYZ Corp. for damages by proving all of the required elements of negligence:

1. Duty of care. As a software developer selling AI-based navigation technology, the corporation owes a duty of care to consumers and the general public to design and maintain a reasonably safe product. (Civil Code Section1714 (a); Rowland v. Christian (1968) 69 Cal.2d 108, 112-13.)
2. Breach of duty. XYZ Corp. breached its duty by failing to properly test its AI system under foreseeable real-world conditions, such as road closures, before deployment. A reasonable software company would have conducted rigorous testing to prevent foreseeable risks. (Mosley v. Arden Farms Co. (1945) 26 Cal.2d 213, 216.)
3. Causation. XYZ’s breach was a substantial factor in causing the harm. (Mitchell v. Gonzales (1991) 54 Cal.3d 1041, 1052.) Here, the accident would not have occurred but for XYZ’s defective software, making XYZ’s negligence the actual cause. Additionally, the harm was foreseeable because navigation software must account for common road conditions like closures, establishing proximate cause.
4. Damages. The pedestrian suffered severe injuries, resulting in medical expenses, lost wages, and pain and suffering, which are compensable under California law.
5. Defenses and comparative negligence. XYZ Corp. may argue that the vehicle manufacturer also contributed to the harm by failing to include a manual override system. Under California’s pure comparative negligence rule (Li v. Yellow Cab Co. (1975) 13 Cal.3d 804, 828-29), the pedestrian’s recovery would be reduced in proportion to any fault assigned to other parties but would not be entirely barred, unless the AI’s autonomy is deemed an independent cause—a defense courts currently weigh on a case-by-case.

This example shows how the existing negligence framework could address AI harm by focusing on human oversight, but AI’s autonomy might complicate causation, prompting defendants to argue it acted unforeseeably, a contention this bill seeks to foreclose.

Strict liability standard. Under California law, strict liability applies in cases where a defendant is held liable for harm regardless of intent or negligence. This doctrine primarily applies to product liability (Greenman v. Yuba Power Products, Inc. (1963) 59 Cal.2d 57, 62). In product liability cases, a manufacturer, distributor, or retailer is strictly liable when a product contains a manufacturing defect, design defect, or failure to warn of a known danger. (Barker v. Lull Engineering Co. (1978) 20 Cal.3d 413, 428.)

Under a strict liability framework, XYZ Corp. would be liable if the plaintiff showed that: (1) XYZ was a commercial supplier; (2) the AI’s inability to handle closures is a design defect under applicable consumer expectation or risk-benefit tests (Soule v. General Motors Corp. (1994) 8 Cal. 4th 548, 566-68); (3) the defect caused the crash; and (4) damages resulted.

The success of this strict liability argument faces significant hurdles, only one of which is addressed by AB 316. First, it is unclear that AI qualifies as a product for purposes of applying a strict liability framework. AI’s status as a “product” is debatable. There are current cases pending asserting strict products liability claims against AI developers that might clarify whether this legal framework applies to AI. For example, in a case filed in the Middle District of Florida, a minor’s parents alleged that Character.AI’s chatbot, marketed as a conversational AI, contributed to a minor’s suicide due to its design and lack of warnings. Character.AI moved to dismiss the case, alleging among things, that Character.AI is a service and not a product, and that intangible content, like it provides, is not subject to products liability. It did not raise an “autonomy defense.” (https://www.cnn.com/2024/10/30/tech/teen-suicide-character-ai-lawsuit/index.html.) The motion to dismiss is still pending and it is unclear which way the court will rule on this important question. Nothing in AB 316 addresses this (AI as a product) issue and whether strict products liability should apply.

A second possible hurdle to applying a strict products liability framework to AI is the question of autonomous conduct. An autonomy defense could prevent liability by arguing that the AI’s independent actions—not a defect—caused the harm, severing the causal link required under strict liability.

It is important to note that the Committee is unaware of any cases in the United States where an AI autonomous defense has been raised—let alone successfully—to avoid liability by AI developers or users. The Committee is aware of a single instance where such an argument was (unsuccessfully) attempted. In a case decided in February 2024 in Canada’s Civil Resolution Tribunal — Moffatt v. Air Canada — a chatbot on Air Canada’s website inaccurately told a customer he qualified for a special airfare the airline did not actually offer. The customer filed a claim after he purchased a ticket, and the airline rejected his request to honor the chatbot’s fare. The court summarily rejected the airline’s “remarkable submission” that implied that the chatbot was a “separate legal entity” and that Air Canada “cannot be held liable for information provided by one of its agents, servants, or representatives — including a chatbot.” The tribunal enforced the chatbot’s promise, ruling that Air Canada was liable for negligent misrepresentation by breaching its duty of “reasonable care to ensure its chatbot was accurate.” (Moffatt v. Air Canada (2024 BCCRT 149, available at https://decisions.civilresolutionbc.ca/crt/crtd/en/item/525448/index.do.) “[I]t should be obvious to [the airline],” the tribunal explained, “that it is responsible for all the information on its website.” (Id.) This bill seeks to codify that “obvious” conclusion—AI developers and users are responsible even for the “autonomous” decisions made by AI.

This bill is largely preemptive. It seeks to eliminate autonomy as a shield, streamlining causation across torts like defamation, malpractice, and products liability. As noted above, this “autonomy defense” has not been legally tested. Defendants facing liability for AI-related harm have instead challenged negligence and strict liability in other ways, including by raising a First Amendment defense, Section 230 preemption, and failure to meet the elements of negligence/strict liability, depending on the case. This bill merely eliminates one other possible avenue of defense. By barring autonomy defenses before they are tested in court, AB 316 ensures AI outputs are legally tied to developers or users, preventing them from pointing the finger at AI’s so-called autonomous conduct as an “independent cause.” For example, in the XYZ Corp. navigation hypothetical above, under AB 316, the pedestrian’s claim would succeed if XYZ’s design or testing flaws are shown, regardless of the AI’s “decision” to accelerate. AB 316 strengthens California’s tort framework for AI harm but doesn’t fully resolve its complexities, marking a bold yet incomplete step in the evolving AI accountability debate.

ARGUMENTS IN SUPPORT: The Organization for Social Media Safety explains its support for AB 316:

Given both the alarming speed at which AI-based products are being deployed and the clear, convincing proof that these products can cause severe harm, especially to children, we must ensure that our standard product liability framework functions as expected to protect consumers. This established jurisprudence has been instrumental in ensuring that California’s products have an outstanding safety record, preventing deaths and injuries for millions of consumers while reliably fostering innovation.

We cannot afford to wait decades for litigation to unfold while Big Social advances novel legal theories arguing that autonomously operating AI, rather than the companies themselves, should bear responsibility for the harm caused. At a minimum, this ambiguity must be clarified now.

Your AB 316 (Krell) does just that. It codifies a straightforward principle: those who profit from the development and use of AI cannot evade liability by blaming the intended autonomy of their own machines. If Big Social wishes to profit from this technology, it must do so within the framework of existing tort law, which incentivizes the consideration of safety in product development and deployment. In this way, we can protect against more broken children.

The University of San Diego’s Children’s Advocacy Institute explains the urgency of this measure, against a speedily changing backdrop:

[W]e cannot wait for the most basic legal question about autonomously operating AI – who is legally responsible if it harms someone -- to be clarified definitively through decades of litigation. It took 25 years for courts finally to decree that Amazon’s marketplace was subject to the same products liability laws as a corner store selling the identical product. It has taken 20 years for courts to figure out how to apply basic negligence and public nuisance law to the harms caused by these companies; a process that is still evolving. According to the most recent data available, the median time frame for resolving just the appeal of a case in California (not including years for the trial) is about a year and a half. That’s the median; by definition half of the appeals take longer.

Hence, AB 316 (Krell) simply codifies what should be obvious: that the persons who profit from the development and use of AI cannot, if the AI harms someone or violates a law, point a blaming finger at the intended autonomy of their own machine in an effort to escape liability.

ARGUMENTS IN OPPOSITION: TechNet, a network of CEOs and senior executives from tech industry, raises the following concerns:

We are concerned about strict liability that this bill would create. The bill disallows the defense that “artificial autonomously caused the harm.” It eliminates key arguments regarding foreseeability, design diligence, and user error from the conversation. This would risk imposing a strict liability for any harm allegedly associated with AI and open the door to frivolous lawsuits claiming the alleged fraud. Additionally, the bill does no demonstrate how courts should evaluate complex questions regarding causation, risk foreseeability, or comparative fault when multiple parties are involved in an AI system as developers and users are included in the same liability. As industry and policymakers alike continue to pursue the responsible deployment of AI systems, it is necessary to recognize that the technology will evolve considerably over time. There are multiple, ongoing efforts within the industry as well as national entities to leverage safety tools and create national and international transparency standards to create a cohesive framework that prioritizes responsible AI deployment. Companies are implementing safety frameworks and developing AI-based detection technologies to monitor other AI systems.

REGISTERED SUPPORT / OPPOSITION:

Support

3strands Global Foundation

California Civil Liberties Advocacy

California Federation of Labor Unions, Afl-cio

California Initiative for Technology & Democracy, a Project of California Common CAUSE

California Labor Federation, Afl-cio

California Nurses Association

Childrens Advocacy Institute

Consumer Attorneys of California

Economic Security California Action

Fairplay

National Ai Youth Council

Oakland Privacy

Organization for Social Media Safety

Techequity Action

The Center for Ai and Digital Policy

Opposition

Technet-technology Network

Analysis Prepared by: Shiran Zohar / JUD. / (916) 319-2334