Senate Standing Committee on Elections and Constitutional Amendments

Okay, we are officially coming to order. The Senate Committee on Elections and Constitutional Amendments, and the Assembly Elections Committee is now in session. Good morning. Welcome to our informational hearing titled Artificial Intelligence and Elections: Protecting Democracy in the Digital Era.

Given the timeliness and significance of today's discussion, we have made this a Joint Hearing between the Senate and the Assembly, and I'm very happy to have the Chair of Assembly Elections next to me here. It's rare that we have these Joint Committee hearings, so I'm excited that we were able to pull this together. And both of us are, of course, relatively new in our second year together in the Assembly and the Senate. So it's great to be working together on this important topic.

I also just want to mention that just yesterday, the federal Senate, the United States Senate, had a statement or an interview with the leader of the Senate Intelligence Committee. It's Senator Mark Warner, and he's from Virginia. And he said that he believes that the United States may be more vulnerable to foreign disinformation aimed at influencing voters and undermining democracy than it was before the 2020 election. And he said that this is due to several factors.

Those are improved disinformation tactics by Russia and China, the rise of domestic candidates and groups who are themselves willing to spread disinformation, and the arrival of artificial intelligence programs that allow the rapid creation of images, audio and video. That's difficult to tell from the real thing. He also mentioned the change from tech companies.

So tech companies rolling back their efforts to protect users from misinformation, even as the government's own attempts to combat the problem have become mired in debates about surveillance and censorship. And this is an Associated Press article that was in many newspapers this morning. So I just wanted to look, lay that out, because I think that this is really important and topical for us right now.

So we know that technology is changing more rapidly than ever. And we're, of course, grateful for the role that technology plays in our lives. Technology in many ways, has made life more convenient and efficient and, of course, has saved lives in many contexts. But we also recognize the pivotal role that technology plays in shaping our democratic processes, particularly in the context of elections.

One of the reasons why we're all here on these committees is because we care about free and fair elections and the democratic process. Rapidly changing technology like artificial intelligence and what we've come to know as deepfakes present real challenges to the integrity and security of electoral systems, not only statewide but worldwide. AI tools are becoming commonplace.

I've used it, my family has used it, and actually, we used it to to come up with the title of this hearing. We use chat GPT to come up for the title of this hearing, and we know that AI brings things to the masses that open us up to what could become improper or unethical acts. So at the forefront of our discussion today is the emergence of artificial intelligence technology, specifically on our elections. When used correctly and safely, artificial intelligence has the opportunity to positively change the world, but if used incorrectly, may have harrowing impacts on society.

When it comes to elections, should AI be used wrongly, it could enable the creation of highly realistic yet fabricated audio visual content. These deepfakes pose significant risks to the integrity of elections and our democratic discourse. The proliferation of deepfake videos has the potential to undermine trust in political institutions, manipulate public opinion, and sow discord among citizens.

Moreover, the use of AI algorithms in electoral processes raise critical questions about transparency, accountability, and fairness. As AI systems are increasingly deployed for voter engagement, predictive analytics, and electoral administration, we must ensure that these technologies are implemented in a manner that upholds Democratic values, protects voter rights, and preserves the integrity of our elections.

Today's hearing provides a platform for experts, policymakers, and stakeholders to engage in a constructive dialogue on the multifaceted challenges and opportunities posed by AI and elections. Before I hand it to my colleague, the Chair of Assembly Elections, Assemblymember Pellerin, I also want to mention that we have panelists who are participating via Zoom today.

So for our remote panelists, please mute your phones or computers, and then please select unmute before you begin speaking. Our IT personnel will put you back on mute when you are done. Once recognized to speak, please make sure you can be seen on the screen. State your name and then you are ready to address the Committee. For today's hearing. We'll be hearing from all the panelists on the agenda prior to taking any public comment. And once we have heard all of the witnesses, we will have a public comment period for those who wish to comment. So with that, Co-Chair Pellerin, I would like to offer it to you to make opening remarks as well.

Thank you so much. Good morning. I also want to welcome everyone to this joint informational hearing, and I want to thank Chair Blakespear and her Committee staff for proposing and organizing today's hearing. I also want to thank our Assembly Committee staff, Nicole Becker and Ethan Jones, who are here today. We have just begun a monumental election year. We do so with our democracy already weakened, with trust in institutions and media at all time lows, the truth is under assault, and with deep voter participation disparities, leaving certain communities out of our electorate.

In this challenging landscape, we confront our first AI election, where generative AI deepfakes will flood political discourse, leaving voters uncertain about the authenticity of images, audio and videos. Easy to access tools empower campaigns, conspiracy theorists, foreign entities, and online trolls to deceive and destabilize our information networks.

Instances of generative AI deepfakes and disinformation have already influenced elections globally, including in Bangladesh, Slovakia, and domestically in Chicago and elsewhere. The American public remains ill prepared for this evolving threat. Here's what we know about AI. The technology is changing rapidly. This is the first tool in human history that can make decisions by itself.

Remember, 'open the pod bay door, HAL'. This is the first tool in human history that can create new ideas by itself. While we are learning to use AI, it is learning to use us. Its potential to disrupt human society is immense. From economics to democracy. We have three bills moving through the Legislature to address the issues and impacts of AI on our elections and our democracy. I look forward to hearing from our distinguished guests today to learn more about AI and the steps we can all take to protect our democracy from the threats that the AI technology may present.

Thank you. And do any of our colleagues have any opening comments that they would like to make? Yes. Senator Newman.

Very briefly, I want to thank both the Chairs for their leadership on this issue and the staffs for pulling this together and thank the witnesses in advance for your participation. This is an important issue. I was very excited about it before I heard Chair Pellerin's dystopian explanation about the hellish world we're about to enter. But important conversation. This is a great first step. Looking forward to hearing from everybody.

Thank you. Anybody else? No? Okay. Assemblymember Essayli.

Thank you. I also appreciate the opportunity to be here. I do agree with my colleagues that AI does pose significant concerns, threats, and I look forward to hearing from our panelists. But one thing as we progress in this conversation, I would also like to hear, and hopefully some of the speakers can touch on, is what are the First Amendment implications on proposed government regulations to AI? Because I think that's always a concern.

If government becomes the arbiter of what can and can't be said or truth, that obviously is going to come into conflict with some of our constitutional rights. So that's sort of where my mind is at and sort of some of the tensions I think we'll be tackling and grappling with as we go down that path. So thank you.

Thank you. Great point. Okay, so our first speaker is Liz Howard via Zoom. Miss Howard is Deputy Director of Elections and Government Programs for the Brennan Center for Justice at NYU School of Law. Miss Howard, welcome to the Committee.

Thank you, Chairwoman. Chairwoman Blakespear, Chair Pellerin, Members of the Senate Committee on Elections and Constitutional Amendments, and the Committee on Elections, thank you for the opportunity to testify at today's important hearing. Generative AI has added a new and complex dimension to the existing threats to to election offices and election vendors.

It excels at imitating authoritative sources, making it easier to deceive specific individuals or the General public by impersonating election officials or forging official election documents. And it can do so on a massive scale. As Howard University Professor Bruce Schneier noted, 'artificial intelligence will increase the speed, scale, scope, and sophistication of threats to our democracy'.

Put another way, many of the threats are not new, but they could become more dangerous in the rapidly evolving AI environment. For the 2024 election, the real challenge is that AI provides agitators new tools to increase the scale of such attacks at little cost and in more sophisticated ways than we've previously seen. This means, in addition to their other efforts to administer a safe and secure election in 2024, election officials are learning to identify, prepare for, and respond to AI related attacks. Of course, mere awareness of the ways that AI might threaten elections is no substitute for actually seeing how it could do so.

Yes.

As part of our work on AI and elections, the Brennan Center has partnered with election officials across the country to conduct tabletop exercises. These are crisis scenario planning exercises that bring together various government officials and others to practice coordinated responses to simulated challenges. In these exercises, we use real AI generated content, some of which I'll show you now, I'm assuming you can see my slide deck?

Okay, great. So first, I'll let Arizona Secretary of State Adrian Fontez provide an overview of some of the concerns that election officials have about AI related threats.

Hello. This is a public service advisory featuring an AI version of Arizona Secretary of State Adrian Fontes. This video was produced as part of our 2024 statewide election security table.

Can we turn up the volume, please?

It was created with both the consent and cooperation of the real secretary Fontas, which, again, is not me. I'm an AI impersonation of him. I want to warn you not just about the quality of deepfakes and how they can be weaponized against the American population, but specifically how malicious actors can try to confuse the public by impersonating our websites.

Imagine a video like this, except it's a hacker who uses my likeness to advertise a fake version of the 'azsos.gov' website to try and trick our voters with a variety of techniques. There are tools that allow a hacker to easily download the content of a website. This allows an attacker to stand up a fake version of the website that looks exactly like the legitimate website. That's why it's so important to look for the .gov domain. If it says '.com', '.org', '.net', well, anyone can register those sites. Some examples of what a hacker could do include showing fake election results on their fake page.

They could also try to trick you into checking your voter registration on this fake site and steal your information. They could also post offensive, racist, false, and or inflammatory language on this fake site and erode the confidence of voters that our elections are being run with integrity. So I'll leave you with these two things.

First, if you're looking for a government website, look for that '.gov'. Not all government websites use it, but most do. And the Arizona Secretary of State certainly does 'azsos.gov'. We also use 'arizona.vote' and redirect it to our main '.gov' website. Second, deepfakes are now almost imperceptible from reality. If you see something hard to believe, that might be a good sign that it's not real. Take a moment to look at the original source and ask yourself, does this seem legitimate, or is it an impersonation?

Well, that video was created with the assistance of our colleagues at the Institute for the Future. I also wanted to show you a deepfake that was created over the weekend by someone who does not have a technology background.

I'm AI Liz Howard. The real Liz Howard created this deepfake using a cheap online tool. This tool could be used by bad actors who pull publicly available video of an election official, say, from a legislative hearing, to train an avatar like me, who provides false information designed to trick voters. [Speaking Chinese].

For those of you like me, the real Liz Howard, who don't speak Chinese, my avatar just warned that these deepfakes could be targeted at Members of a specific party, gender, race, or ethnicity. And while these deepfake videos received a lion's share of the public's attention when it comes to AI, many election officials are concerned by the less flashy, but no less potentially disruptive uses of AI, many of which AI Secretary Fontes spoke about. For example, easily accessible AI tools like chat. Oh, sorry.

On this slide, you'll see an innocuous looking image of a map with drop pins indicating absentee ballot drop box locations. AI images such as this one can be easily generated even in response to simple text message prompts, and these images can be used to disrupt the election in multiple ways.

For example, this map with false information could be quickly spread across large networks, even by well intentioned community Members via WhatsApp or other messaging services. And on this next slide you'll see another example of one of the concerns that AI Secretary Fontes mentioned.

Easily accessible AI tools like chat GPT can be used to copy election office websites or their official social media accounts and provide false information about elections. Here you could see a social media post by a spoofed election official account falsely informing the public that ICE agents will be at the polls.

You can imagine how disruptive this could be. And here's another example showing a spoofed election official website with false election results. And this one has false election results, but it could have just as easily been a false log of voters indicating how each individual voter voted.

Of course, most of these scenarios we use could have happened without AI, but now agitators can more easily launch attacks on elections on a much larger scale. Not surprisingly, with the widespread availability and low cost of AI, we're starting to see election officials experiment with this technology.

According to the Brennan Center's 2024 survey of local election officials from across the country, 7% of local election officials are currently using AI. Reported usage includes drafting social media content or press releases, locating polling sites, and translating materials into different languages.

13% of local election officials have been approached by a vendor or outside group with products advertised as using artificial intelligence. Also not surprising, in the current environment, in response to a question about whether guidelines by federal, state, or local agencies for the use of AI in elections would be helpful, 33% of local election officials indicated yes, it would be helpful, while 45% responded, I don't know.

In conclusion, for years, experts have been warning about the threats that AI poses to elections, even before recent advancements, including those from misinformation directed at the public, phishing attacks, and other attacks against election infrastructure, many election offices have already implemented significant and successful steps to protect their infrastructure and staff from these threats.

Early conversations and awareness about AI's capabilities and how to prepare for worst case scenarios will help election officials build on their pre existing security plans to prepare for the more sophisticated and widespread attacks that AI may bring.

But of course, to most effectively respond to these threats, election officials will need the assistance and cooperation of other government officials, community organizations, members of the public, and others. Thank you again for the opportunity to join you today. I look forward to your questions.

Thank you so much, Miss Howard. We appreciate seeing that very much. We're going to hold our questions until we have heard from all the panelists. So we will now move on to our second panelist from the California Secretary of State. We would like to welcome Cassandra Patterson to come forward, Acting Deputy Secretary of, sorry, Acting Deputy Secretary of State for Legislative Affairs, and also Mike Somers, who is the Elections Security Program Manager. Thank you for being here today. You may proceed when ready.

Good morning, Madam Chairs and Members of the Committee. Hi, I'm Cassandra Patterson, Acting Deputy Secretary of State for Legislative Affairs. On behalf of Secretary Weber, I would like to thank the committees for the opportunity to appear before you today to discuss the important issues of AI and election security as it relates to protecting our democracy.

The Secretary of State is the Chief Elections Officer of the State and is responsible for statewide election administration, promoting voter registration, encouraging voters to vote, and promoting civic learning and engagement. The Office of Election Cybersecurity, created by AB 3075 in 2018, is a collective effort involving multiple divisions of the Secretary of State's office. It has subject matter experts on elections administration, technology, communications, cybersecurity and physical security. Today, we would like to share information with you on how the SOS is combating the negative use of AI tools by bad actors.

With me today is Mike Somers, the Secretary of State's Election System Security Program Manager, to present the work of the Office of Election Cybersecurity in our efforts to combat misinformation. We also have additional subject matter experts from our office in the audience, should they be needed. We have Jana Lean, Chief of Elections, Patrick Ryan, Chief of Security, Tamara Johnson, Chief of Finance, NaKesha Robinson, Voting Systems Security Director and Joe Kocurek, Communications Director, and I will now turn it over to Mike.

Hi, thank you. Good morning. Thank you for having us here. My name is Mike Somers. I'm a registered election official and have been part of the election office since 2015. The office of Election Cybersecurity was created when the Secretary of State's office worked closely with the Legislature to pass AB 3075 in 2018.

This Bill established the Office of Election Cybersecurity with a mission to prevent or reduce election related cyber incidents and to prevent or respond to misinformation about elections. The office was established in response to the unprecedented levels of misinformation spread during the 2016 presidential election.

In 2017, the Secretary of State's office worked to identify opportunities to better counteract this flood of misinformation and ensure the security of elections. So, following the 2016 presidential election, our office observed a lack of coordination among local, state, and federal officials around the issue of election security and responding to misinformation.

The Secretary realized that more needed to be done to communicate across all levels of government about threats to and misinformation about elections. Since 2019, the Office of Election Cybersecurity has worked tirelessly to combat misinformation and establish strong partnerships with local, state, and federal agencies to protect elections in California from emerging threats.

As you know, election misinformation is not unique to the digital era and has always been part of the elections information landscape. As long as there have been elections, there have been people willing to spread bad information for nefarious purposes. For example, one of the oldest tricks in the book is to tell people your election day isn't on Tuesday, you vote on Wednesday. Other examples might be persons with outstanding credit card debt will be arrested at the polls, or text messages such as vote for your candidate via text message today.

Versions of these messages are still used today. Despite the advantages that come with being connected by social media, we have also witnessed drawbacks. Narratives being pushed at the national level have created confusion among California's voters. These national voices are actively working to convince people that voting system technology is insecure, and they continue to spread wildly inaccurate, false, and misleading information about election technology. In addition, foreign state actors are also able to easily amplify these messages or supply additional misleading content.

Now, this misinformation is particularly frustrating since the Secretary of State's office is legally mandated to certify any voting system prior to its sale and use. California has some of the most strenuous voting system testing and certification programs in the country. New voting systems applying for certification must undergo months of extensive testing for both security and usability, and confirmation that voting systems do not and cannot connect to the Internet. This is just one example of how the information environment around elections is already fraught with incorrect or even intentionally deceptive information.

So as artificial intelligence technology becomes more widely available, we know that AI will create additional challenges for communicating with the public about elections. One way the AI is expected to impact communication about elections is by lowering the barrier of entry for creation of misinformation. The ease with which bad actors can create content will likely make reacting to misinformation even more difficult than it already is. Misinformation spreads faster and further than information that corrects it, and this imbalance will only be exacerbated by the ability to create more content more rapidly.

Unfortunately, it will also likely improve the quality of misinformation in other languages that target non English speakers, and similarly, it may also improve the ability of foreign actors to produce English language misinformation. Another especially concerning threat from AI misinformation is the use of deepfake technology.

The ability of a bad actor to use the likeness and voice of a trusted source of election information, that of an election official, has the potential to widely disrupt communications about critical election information to the public. Now, the good news is that experts in communications and combating misinformation have noted that the strategies already employed by our office can work to prevent misinformation, whether or not it's developed using AI technology. Our key communication goal is to provide high quality and easy to access information to as many voters as possible.

This strategy is often referred to as pre bunking where good information reaches the audience before bad information. This reduces the number of claims that might need to be debunked. Our outreach and education efforts focus on delivering accurate and timely election information as frequently as possible.

Getting accurate information to the public helps inoculate people against misinformation in the first place. Maintaining consistent messaging and accurate information helps combat misinformation directly, and in the case that misinformation becomes viral, we will respond directly to address it.

Our messages focus on voter rights, driving traffic to our website as California's trusted source of elections information, and providing the direct contact information of elections officials, which enables voters to have their questions answered or to report misinformation directly to our office through the voter hotline.

We use a combination of earned and paid media to reach as broad an audience as possible, and we know that our information has excellent reach based on social media analytics. And we aren't alone in our efforts.

We coordinate closely with all 58 counties to provide clear, consistent messaging, and our office provides high quality resources to county elections officials, including social media messaging toolkits for each election, and we ensure this information is available to all voters by translating these messaging materials into nine languages.

We also provide security resource toolkits that help identify no cost security resources for both technical and communications topics, and we provide templates to help develop incident response, crisis communications, and continuity of operations plans.

In addition to coordinating communications efforts around the state, the office has built strong partnerships with local, state and federal agencies to bring a whole of government approach to protecting California's elections from all types of bad actors, and we are confident in our ability to respond to a variety of situations and scenarios.

We partner with the California Office of Emergency Services to coordinate the California Election Security Task Force. The task force has many Members, including CHP FBI, CISA, county elections officials, local law enforcement and more. The task force focuses on three main goals.

The task force has helped to develop and conduct election security training and exercises for local elections officials on many security topics. Participants provide threat intelligence information and briefings geared towards protecting elections. And finally, the task force coordinates plans to secure the conduct of elections in cases of emergency.

Finally, we believe that over the last five years, our efforts and the efforts of our partners have made California a national leader in election security, and through our partnerships, we are well positioned to respond to many types of election security incidents. Combating misinformation and cybersecurity threats is a continuous process. It requires constant action, vigilance and partnership.

Through our ongoing efforts, we will continue to ensure that every voter can cast a ballot free from interference and with exceeding confidence in our elections. Thank you so much.

Well, thank you very much. We appreciate the information. Our third presenter, we have four total presenters. Our third presenter is Becky Waite and she is joining us via Zoom. She is the head of global response for OpenAI. Miss Waite, welcome to the Committee.

Hello. Can you hear me okay?

Yes, we can.

Great. First of all, I just want to thank the Joint Committee for the invitation to present on this important topic. My name is Becky Waite and I lead the global response team here at OpenAI. The work includes preparation for global elections around the world in 2024 and beyond.

Previously, I worked at Meta, where I helped direct efforts to prepare for the 2022 and 2020 us elections, and I began my career in both the nonprofit and public sector. Let me just, I'm seeing. Let me just share my screen really quickly so that we can walk you through a quick slide deck.

Okay, I apologize. I don't think I can share on Zoom, so I hope you'll forgive me. I'll just have to walk you through verbally. So quickly, we are thinking the remit of our team is global elections in 2024 across the world.

But we are obviously gearing up for the US 2024 elections and our approach is really to continue our platform safety work by enforcing measured policies, elevating accurate voting information, and improving transparency.

We have a cross-functional effort at the company dedicated to election work, bringing together expertise from across a range of groups, including our safety product teams, threat intelligence, legal, engineering, and policy teams, to make sure that we can quickly investigate and address potential abuse and identify ways that we can help elevate positive information around the elections.

Before I dig in, I think it might be helpful to orient around how OpenAI operates. So we were really started as a nonprofit, and that nonprofit still governs us.

We are really driven by our mission to build artificial general intelligence, that is AI, that is at least as smart as humans, to be safe and beneficial for all of humanity. That is still something that drives all of our work today. We have a few different products and services that are broadly available. The first is Chat GPT.

It's something that you all might be familiar with. It's a chatbot that helps people interact in a conversational like manner. We also have, we have that available for free for people to broadly use, and then we also have a premium version of that that is paid. In addition to that, we have something called our API.

It's for developers and allows them to engage and build on top of our, our services and tools for their own uses. We also have two products that are in research phase and are not broadly available for public consumption, public use or developers, including a synthetic voice and video model.

Those are still in research testing, and we have not developed those for public use at this time. We want to make sure that when we, if and when we do deploys that deploy those, we're doing it really safely. That's top of mind for us. We design our models with principles and objectivity in mind.

We recently released something called the model spec that outlines how we think about public principles for developing our models. And we have a series of safety efforts in place to make sure that as we bring new models into the world, that we really robustly test them.

Something that we do is called red teaming, which might be very familiar to this group, but effectively gets a group of experts to try and break our model, identify ways that it might be abused, and allow us to patch those concerns before releasing it broadly. This is a really iterative process.

So as we discover new issues in the model, we patch it before public release. So talking about sort of our safety stack for the elections, we have a three pillar model. The first is really around preventing abuse. What does that mean?

Making sure that we have really good policies in place, making sure we have good monitoring and defense mechanisms in place. On the usage policy front, we have a couple of policies that are particularly relevant for the election space. The first is we don't allow impersonation on our tools and services.

This makes sure that we don't have folks impersonating election officials or public services or other sorts of public institutions that might be confusing for voters in the election context. We also don't allow the use of our services or to build on top of our services to discourage people from participating in Democratic processes.

I know one of the previous panelists mentioned some of the ways that folks might get discouraged by saying that it's an incorrect date time, confusing people about eligibility criteria, misleading people about eligibility criteria. We prohibit the use of our tools for that kind of information.

And then finally, out of an abundance of caution, we do not allow the use of our tools for or political campaigning or lobbying. We have a broad set of safety systems that are really built from the bottom up in terms of our tools to detect unwanted content.

We also have a reporting mechanism for users to report potential violations on GPTs. And finally, we have an entire investigations team that reviews, looks for, and takes down accounts associated with any kind of influence operations. We just put out a report earlier this week or late last week announcing takedowns of several influence operations.

I will say to date it does not seem like there is really extensive activity, but we are ever vigilant looking for potential activity and making sure that we're trying to catch those accounts and removing them.

Something that we've heard in this briefing to date and from our conversations with policymakers is that deepfakes are very top of mind for folks in the 2024 election globally, and we have invested significant time and energy in figuring out how to address the risk of potential deepfakes.

First, Dall-E, which is our image generation model, has guardrails to prevent people from creating images of real people, including candidates. These are pretty robust guardrails, so if you go in and try to create an image of a real person, it will refuse.

We also believe that work in provenance, meaning the ability to identify the origin of a piece of content, is important when thinking about the potential harm of deepfakes, particularly images that might appear innocuous but taken out of context could be harmful. And we have a few elements of work around provenance that we have invested in.

First, we joined the steering Committee of C2PA, which is the Coalition for Content Provenance and Authenticity, and implemented their standards on our images. We also recently began providing researchers with early access to a new tool that will help identify images created by OpenAI's image generation tool.

And finally, we've added watermarking to a couple of our models as well, which is a different methodology of provenance. Finally, we want to talk about sort of safety by collaboration, so we frequently publish our research in AI safety, outlining both our approach and key findings from our teams.

To that end, we have launched a grant program to provide $10 million for technical safety research, tackling issues from weak, strong generalization, interpretability and scalable insight, and more, all of which are critical for us to figure out as we continue to build these models.

Finally, we recently announced the creation of the Societal Resilience Fund in partnership with Microsoft. This follows from the Munich Tech Accord that was signed in February or March by many of the industry companies, and our Fund has $2 million, which will provide AI literacy to election authorities and civil society in key regions around the world.

With that, I will pass it off. Thank you so much.

Well, thank you very much Ms. Waite. We appreciate your sharing that information and the Committee Members have your PowerPoint in their folders so they are able to see that. Our fourth presenter is David Evan Harris. He is the senior policy advisor for the California Institute for Technology and Democracy. He is also appearing via Zoom.

So welcome Mr. Harris.

Thank you. Chairs Blakespear and Pellerin and Members of the Committees. It's an honor to be able to speak with you today. I'm sorry that I can't be there in person. I'm recovering from COVID and still testing positive and didn't want to get you all sick.

I thank you for this invitation and for yourselves choosing to focus your public service on such a critically important topic as elections. As introduced, I'm David Harris and I'm here in my capacity as a senior policy advisor for CITD, the California Initiative for Technology and Democracy. CITD, the newest project of California Common Cause, came together to seek state-level solutions to the threats that disinformation, artificial intelligence, deepfakes, and other emerging technologies pose to our democracy and our elections.

I'm also chancellor's public scholar at UC Berkeley and have taught at the Haas School of Business since 2015. With classes including civic technology and AI ethics for leaders. I teach my students about all the ways that technology plays a role in our democracy.

I also previously worked for close to five years on civic integrity, misinformation, and responsible AI at Facebook and Meta, where I was incidentally a colleague of the previous speaker, Becky. It's also relevant to mention that I'm a senior advisor for AI and elections to the Brennan Center and have been a close collaborator with our other previous speaker, Liz Howard, for much of the past year.

I've also advised the European Union, the White House, NATO, and the United nations about AI, democracy and disinformation. But your jurisdiction, California, and these two committees in particular, are really the center of the universe as far as I'm concerned right now.

AI is a powerful tool that can bring great benefits to people, but it also, if not governed judiciously, has the potential to cause great harm. This is why so many AI company CEO's have themselves called for regulation of AI.

And that's why I personally have chosen to work with CITD because I believe that legislative solutions are the only way that we will get a meaningful grip on the scourge of misinformation AI and their impact on elections. We simply can't count on Silicon Valley companies alone to do what is right for democracy.

There are a few reasons market pressures are driving social media companies to make dramatic cuts to election integrity teams and content moderation more broadly. Elon Musk cut Twitter's staff by 81%, and Mark Zuckerberg said that he admired what a lean company Elon had made Twitter before then, cutting tens of thousands of Facebook workers.

Social media companies see that they can now get away with deprioritizing their efforts to fight myths and disinformation. I applaud Chair Blakespeare's sighting of Mark Warner at the introduction to this hearing in saying that we're less prepared today than we were in 2020, I completely agree.

Social media platforms are poorly staffed, and some have simply abdicated responsibility for taking elections seriously. AI as well, is a race to the bottom, with trusted companies like Google now telling us to eat a rock a day and put glue on our pizzas to keep the cheese sticking on them.

I live in San Francisco, and something I've now heard twice from different friends in the startup community is that tech entrepreneurs are openly talking about the need to build AI products fast before they're illegal. Yes, you heard me correctly.

People in the tech industry here in the Bay Area are talking about the need to build the AI products quickly before they're illegal. This includes products that produce deep fake images, voices, and videos that can undermine our elections. Yes, you heard that right. Silicon Valley wants to build fast before our laws can catch up.

And that's where you come in. But first, let's talk about Washington. More than one Member of Congress in Washington has said something that irks me immensely, a version of the statement that we failed to regulate social media and we can't fail again with AI. I refuse to accept that idea.

The truth is not that we failed to regulate social media as a country, but that we are actively continuing to fail every single day to regulate social media. Washington is likely to keep failing to take meaningful legislative action on both social media and AI regulation. Europe is trying hard, but they can't do it alone.

And while they're actively investigating Facebook and Instagram for allowing Russia to buy ads subverting the upcoming European parliamentary elections, they have no jurisdiction here. That is why California is so crucial. Many of you have heard of the California effect and the Brussels effect.

These are two political science terms for the extraterritorial impacts of laws passed here and in Brussels. This applies to laws about the environment, about privacy protection, and after this legislative session, I believe it will apply about elections, AI, and social media.

In your background materials for today's hearing, I was pleased, excuse me, I was pleased to find summaries of two of the bills from the cited package. The first is AB 2655 by Assemblymember Berman.

This is the deepfake labeling on social media bill that requires social media platforms to label election-related deepfakes, that deceive voters about candidates and election officials, and in the most egregious cases, ban them for a limited period of time close to election day.

The second comes from Chair Pellerin, AB 2839 the deepfake free campaigning close to elections bill that bans online, or that bans offline election-related deepfakes that deceive voters about candidates and election officials in political mailers, robocalls, and TV ads for a limited period of time close to election day.

Another bill in the cited package that I'd like to call your attention to is Assemblymember Wick's AB 3211, the California provenance, authenticity, and watermarking standards. This is not an election-specific Bill, and it won't be coming to your Committees.

But it is a critical technological tool that would facilitate the ability of tech companies to comply with AB 2655 and 2839. It requires AI companies to insert invisible, difficult to remove watermarks into both AI generated content and authentic content.

I was also very happy to see in your briefing materials a reference to the February 2024 AI elections accord signed in Munich by 20 major AI companies. The companies made commitments in that statement that most of them have failed to live up to, including on applying watermarks to AI generated content.

Many of the same companies made similar commitments in last July's White House voluntary AI commitments. These types of declarative declarations and voluntary commitments are a joke.

When there is no timeline for compliance and no accountability mechanism, we must assume that tech companies will continue to predictably fail to defend our democracy when it is not legally required of them. I will note that OpenAI, represented here today by Becky Waite, is an exception in that they have applied. They have taken seriously some of the commitments that they made in those accords.

But when only one or a couple of companies choose to comply, it leaves the entire ecosystem vulnerable and drives abusers to simply use the other readily available products on the market to subvert elections.

As authors of two of the bills in the cited package, Chair Pellerin and Assemblymember Berman present here today deserve to be honored already for their work protecting elections. I hope that in future sessions, others among you will consider working with cited on additional bills as our work will surely not be done this session.

In closing, I want to remind you that you are standing on the front lines of democracy. Not for your district, not for your state, for the world. Thank you for your critical work to protect our elections.

Well, thank you Mr. Harris, that was very interesting. I would like to invite the Secretary of State representatives to come back up here. I have a couple questions before turning it over to the Committee. And also I know that you noted this Mr. Harris, but both of those authors are sitting here today with us.

So we're excited that Chair Pellerin and Assemblymember Berman are here and could offer any additional thoughts that they have. But I'll just start. I wanted to ask you, Mr. Harris, about, well, you said a number of things that I thought were very interesting.

So one of them was your irritation at the phrase that we're actively failing to regulate social media. I think one of the reasons that people say this again and again, well, we shouldn't let this happen with AI, is that it's what becomes normalized.

So even how do you start to take away people's uses and an expectation of what types of either time constraints or content constraints, time, place and manner restraints, any type of constrain on how people use social media? I think it's the same idea with AI.

And so the phrase that was used by the Secretary of State, lowering the barrier to entry, I think is a really great point because people with very little tech savvy can now do things that they couldn't before. So, but it hasn't been widely adopted yet.

So the question is, what kind of legislation or regulations do we want here in the State of California, the fifth largest economy in the world, home of Silicon Valley? And so I wanted to ask that direct question of you, Mr. Harris. Do you have legislative solutions that you did not cite?

Not the existing bills, but other ideas that you think would be most impactful or relevant, and particularly keeping in mind that important balance with the First Amendment and the concern about government overreach and also government being involved in political speech at all.

So I just wanted to see if you had any direct suggestions for us here in the Legislature.

Absolutely. Thank you so much, Chair Blakespear, for the question. We're actually currently in discussions on the California Initiative for Technology and Democracy team about what bills we might be able to sponsor or what ideas we might be able to bring to potential authors for the next legislative session.

So I think it would be a little bit too early for me to speak about specific bills that we might introduce, in addition to those that I mentioned, however, I recently facilitated a panel about elections and social media and AI with some senior officials from the European Union who were visiting San Francisco about a law that they have in Europe called the Digital Services Act.

And the Digital Services Act came fully into force this year in February, and it has some very specific guidelines that were added on to the act about the protection of elections.

And I think that the Digital Services Act is a really interesting model for legislation around the protection of election integrity because it does not impinge on freedom of expression.

In fact, if you look at the Digital Services Act, it mentions many, many times throughout that it does not cut off anyone's freedom of expression and that its provisions should not be applied in ways that impinge upon that.

They don't have the First Amendment in Europe, but they do have fundamental rights, of which freedom of expression is one.

What they do do is that they require all social media platforms and search engines to conduct rigorous risk assessments of their social, of their platforms, of their technologies to see what impact and what risks their platforms pose to elections and democracy.

Additionally, they are required to conduct those same kinds of risk assessments to assess potential risk to mental health, to children, to public health. And then after those risk assessments are produced, the companies have to propose risk mitigation strategies. They have to have third-party auditors audit their risk assessments and mitigation strategies.

And they also have to allow independent researchers, academics like myself, or people from independent research organizations to have data access to check and see how effective their efforts to protect democracy and elections are.

So I believe that there are a number of provisions in the European Union's Digital Services Act and also the EUAI Act that could serve as inspirations for future legislation in California.

Okay, that's interesting to hear. Thank you. I wanted to ask a question also of the Secretary of State. So I understand that the public can email information, can email you about information that is false or misleading and that the Secretary of State will look into it and hopefully do something about it.

And so the email is votesher s u r e, votesure@sos.ca.gov. And I'm curious about how many social media posts are reported to the Secretary of State, how much follow-up you're doing. What exactly is happening when people do email that?

And I, you know, I think of different areas where the consequences are do actually have a chilling effect on behavior we don't want. And so, and things like, I mean, the severity of consequences around, like corruption or is something that comes to mind or like copyright infringement.

You know, we have a lot of regulations and consequences around different areas of law that have to do with data and facts and how we present them and if they're misused. And so I'm wondering about the Secretary of State's involvement in that part and how robust it is.

Yeah. Thank you. That's a very excellent question. I would say vote sure is just one of the tools that we have to collect information we hear from the public through multiple channels. Our voter hotline is actually probably more widely used than some of our email addresses. We also have,

Maybe you could say the voter hotline number out loud,

1-800-345-VOTE. But we have an elections email as well. You can go to the elections website and email us from there. So we get information from the public through multiple different channels. I would say reporting of misinformation is not as high as you might think in terms of the overall volume.

Most folks are calling for, where's my polling place? Or I'm having a challenge voting, those kinds of things. We're always solving those kinds of problems as well. But when we do get reports of misinformation, we certainly take a look at it. Our communications team focuses again on looking at what is the misinformation. Our evaluation is mostly about.

Are they providing incorrect information. Information. Again, things like election day is on Wednesday. Those are the kinds of misinformation that we are capable of, you know, working on legal challenges to. But, you know, speeches is definitely an important part of the political landscape. And so I think when you're talking about combating misinformation, speeches is certainly a challenge.

And so our legal team focuses on evaluating misinformation that's reported to us to make sure that we're not infringing on First Amendment speech. And then we engage with partners as well.

We work with National Association of Secretary of State and the National Association of State Election Directors to engage again across the country with social media vendors as well and having conversations with them about our concerns. So, you know, we try to tackle misinformation from multiple angles through multiple partnerships.

But so just so I can get a better understanding of the specific question, like, if there is a social media post that puts someone's photo and hate speech speech together with it, and someone reports that to the hotline or to the email, are you following up on that?

Do you refer that over to law enforcement or what exactly happens?

It depends on what the content of the post is. Right. So that's where our legal team and our communications team work hand in hand to sort of evaluate the misinformation.

If it's misinformation on elections that maybe is spreading, you know, will work on messaging that will, you know, try to put out, you know, direct responses to viral misinformation. But if it's, you know, if it's more threatening, we do work with, we do work with law enforcement.

Okay. Yeah. Would you like to come forward, sir?

Yeah. I'm the Communications Director, and I was.

Yes, go ahead.

I just wanted to say that there are two approaches. One of them is not very effective and the other one is much more effective. The sort of whack-a-mole approach that you're talking about, we will help. We will perhaps forward information like miss and disinformation and consider it.

But it's not a very effective, and as has been alluded to, it's a first amendment, perilous approach to dealing with mis and disinformation.

But there are certain patterns in misinformation that we try to, as, as Mike mentioned earlier, we anticipate and then we will respond in, to inoculate folks so, you know, their date and time, you know, qualifications, your legal rights. There's a, you know, there's a Voter Bill of Rights in California.

All those things we distribute in volume, like he mentioned that we have paid advertising on social media. We send out these messages and we have 41 to 51 million impressions per platform with that. So the AI is something new that we'll have to deal with. But we're not in the business of infringing on people's First Amendment rights.

What we are doing is our business is to ensure that people have accurate information to exercise the franchise. And that's, you know, that's a huge enterprise already.

Yes, I appreciate that. I mean, it's what, the language that he used was very good. It was pre-bunking. So that you're reducing the amount of. What's the other word that's after that? Debunking. You're pre-bunking to avoid debunking. Right, correct. Exactly.

So I appreciate that it does seem like the whack a mole or as the landscape is shifting, that there will need to be, just seems like there would need to be something more robust than that as well. It's certainly really important and I'm glad that you're well-resourced to be able to do that.

But with 58 counties and all this changing technology, it seems like, you know, we'll need more than that.

It's always a challenge. And I think the other interesting thing is you never quite know what the next thing is. Right. The next issue, you know, and so we definitely have to continue to investigate.

Right. Okay, great. Well, thank you. I would like to call on my co-chair here.

My Bill would actually help the situation by banning that digitally created materially deceptive materials 120 days before an election and after an election. So we're hoping that that will go through and would have that injunctive relief tool for people to combat anything that they're seeing.

So I'm just thinking about the creation of the Office of Election Cybersecurity in the Secretary of State's office, which of course was done by our Assemblymember Mark Berman, certainly predates our recent accelerated development of AI tools.

And so I'm just wondering what your capacity is to address these challenges posed by AI and the threats to California elections.

So that's one of the reasons - sorry, Jana Lean, Chief of Elections. It's one of the reasons why we do partner with state and federal agencies to assist us with that.

Of course, resources are always needed to, additional resources are always needed to help with us to develop these type of miss and disinformation campaigns and how to address those and as quickly as possible.

But I do think that we have, through our election security task force, to try to figure out and identify attribution of where they're coming from and then to identify potentially where it is, where is spreading as quickly as possible. I think that will help us combat mis and disinformation as quickly as possible.

Also, we have a lot of channels that Mike was talking about to get information out quickly to county elections officials.

If we're seeing something at the Secretary of State and we've identified that it's mis and disinformation that's out and that could be spread even further within California, we have an ability to do that very quickly, and we've developed that over the years in order to make sure that that information gets out there and to give them true information.

If we have put together a press statement, if we put together a social media post, we will share that on the channels to make sure that it's everywhere and that folks can actually see the true and correct information.

I just want to be on record thanking the Secretary of State. As you know, I worked as a chief election's official for 27 years or so, and just your collaboration with the counties and your ability to work quickly with the counties when something happens has just been admirable. And so, thank you for your leadership there.

The other question I have here is I loved listening to Liz Howard talking about the tabletop exercises and wondering if California is looking at doing something similar to that.

We actually have already done that. That's part of what the office of Election cybersecurity. We've done a lot of training and a lot of tabletops throughout the whole State of California. We offer it to counties specific. We can tailor it to them if there's a specific type of scenario that they want to go over.

We have conducted many trainings, especially during the summertime when we actually have a little bit of time while we're preparing for the next election. We have a whole series. It's a training series, and Mike can go more into that if you'd like to hear about that.

Yeah, I'd like to hear a little bit more.

Yeah, absolutely. Thank you. We've been very excited to provide training opportunities to counties. We've tried to take a number of different training methods and approaches to providing information. We've done some lecture type series on a variety of topics, incident response, crisis communications, continuity of operations.

We've tried to bring in experts from our partner agencies and provide technical and expertise to county elections officials so that they can continue to develop their plans and continue to make those better. We've provided, as you mentioned, tabletop exercises, which are really excellent and sometimes stressful events where, you know, you present them.

You present these county election's officials with some of the scenarios that you can imagine being a problem and being a challenge and asking them to come up, you know, and work through their plans so that they can, you know, effectively test those plans.

We've partnered, so we've conducted tabletop exercises that are produced by our office specifically, but we've also participated in national tabletop exercises through CISA. They run a tabletop; the vote exercises every year. It's a national level exercise, and they have typically gotten 50 state engagement.

And so, it's an excellent opportunity to share information across the country, hear from other elections officials in other states about their activity. And we've, in fact, taken an active role in helping to develop that exercise at the national level. And then we've run very complicated simulation style tabletops where we give them time pressure.

We run basically a simulated election day and kind of throw scenarios at them on a sort of time crunch basis, running, you know, running a full election day in about 90 minutes and putting up fake media, you know, social media, putting up invented news posts and things like that, and then asking them to sort of respond not in real time, of course, but in, you know, under, under time pressure to really sort of hammer home that, you know, we need to make decisions and make phone calls and make those contacts and connections.

So we've done, we've done a lot with the exercises and trainings around the state.

That's great news. Thank you. And I think it really is going to require all of us working together. And this is such a quickly changing landscape, as we all know. And things could have really dramatic effects when they are put out on social media and people believe them.

I know when we went to the cited conference, they showed a false picture of the Pentagon being bombed. And on that day, the stock market just crashed as a result of that one photo that went viral. Actually, if we could bring David back, I have a question for him as well.

I'm thinking you're still hearing me, so I'm going to go ahead and ask?

Yes, I am. Yeah.

So, thanks for all the work you've done working with the European Union and the steps that they've taken to regulate A.I., including efforts to use A.I. to influence elections. But the United States has much different protections, as we know, for the freedom and speech than those that apply to Europe.

And how I know that we're getting questions about this. So, how do the free speech protections in the U.S. constitution create additional challenges for regulating the use of A.I. in attempts to influence elections?

That's a great question. I'll start by the compulsory. I am not a lawyer, I'm a sociologist, but I have spent a lot of time thinking about this issue. So, I think it's actually a popular misconception that Europe's laws about social media impinge on free speech.

And that's because I'll emphasize that the law that I mentioned that Europe's Digital Services act, it does not change any definitions of what speech is illegal or not permitted.

It simply requires that the companies that operate social media platforms and search engines assess the risks that their platforms pose to elections, democracy, public health, mental health, and address those risks in ways of their choosing.

And a lot of the Digital Services Act also rests on something that's called co regulation, and that's built upon something called the Code of Practices on Disinformation, which was actually produced before the Digital Services Act as a collaboration between companies that run online platforms and search engines, and the government of the European Union, the European Commission.

And so, there's this idea that it's a binary thing. It's like either we stay out completely of social media, or if government gets involved some way, all of a sudden, you're cutting off people's freedom of speech. And it's simply not a binary like that.

There are many, many ways that we can Institute laws that allow for us to look closely at the problems that social media platforms are causing for elections and give the owners of those platforms and the operators of those platforms the tools that they need to address those issues.

And also give the public, give researchers access to the information that they need in order to be able to tell if the platforms are actually doing their jobs.

One of the biggest challenges is that these companies are not only unaccountable for protection of elections and democracy in the United States, but they are also so opaque that researchers and outsiders don't have the ability to even know if the companies are doing everything that they should or could be doing.

One example from Europe right now, and if you read this article, that came out a couple months ago about this doppelganger operation. It's a misinformation campaign that's being run in Europe by Russians. It reads like it's 2016, but it's 2024, literally.

The Russian government is buying ads on Facebook and Instagram to promote pro-Russia, anti-Ukraine sentiment in ways that will impact the European parliamentary elections. And it's just, it is so deja vu. It's absurd. And so, I think, in some ways, Mark Warner was understating the idea that we're less prepared.

I think we might have been more prepared in 2016 than we are today for today's election. So, you know, I think, again, the key thing is not buying into this binary that any regulation related to protecting our democracy impinges on freedom of speech. It's just not that, and another element of this, again, not a lawyer, but is the questions around Section 230 and how that evolves. Congress has shown some appetite to make some changes there, but it's not getting very concrete.

Section 230 is also, in some ways, a barrier, and that is a very old law that has a huge implication for what we can do in this country and needs to be rethought.

Thank you. I've got lots of other questions, but I'm going to open it up to my colleagues, who may ask them as well.

Okay, thank you. Assembly Member Berman.

Thank you, Chair Blakespear. And I really want to thank Chair Blakespear and Chair Pellerin for continuing convening this hearing. The idea around the Office of Election Cybersecurity actually came out of a similar type hearing that we had in 2017, I think.

So, I'm optimistic that a lot of good ideas will come out of today, and I appreciate the presenters, appreciate the work that the Secretary of State's Office does.

And it was really kind of fun and informational and enlightening to hear about the work that the Office of Election Cybersecurity is doing, especially with counties, and helping to disperse best practices and make sure that they're getting the resources they need, as opposed to small counties having to go to the office of Homeland Security directly.

So, it's really great. I appreciate the work that you're doing. I also appreciate OpenAI participating today, and I firmly believe that the, the things that a lot of the A.I. companies are saying in terms of wanting to: a. lot of them are actually saying they want regulation, which is very abnormal for tech.

This is sort of the first time that we've seen a big new industry say, no, we actually need regulation. We need government to get involved. Usually there's a strong libertarian bent, and they want government staying as far away from them as possible. And I think a lot of that is very sincere.

But I worry, as, as Mister Harris sort of alluded to, that the pressure for more investment and the pressure for higher valuations is so strong with these companies, especially with smaller companies, but also with big companies like OpenAI and others, that at the end of the day, when you have to make a decision between increasing your security or paying more attention to your risk assessment teams or launching that new product faster and getting more users and increasing your valuation, that they're going to go with the latter.

I saw this when I was a baby lawyer just out of law school in 2008, 2009, working for a big law firm. And one of the first things I did was I worked on creating privacy policies in terms of use for a lot of online companies, especially online video game companies.

And there was this great video of Mark Pincus, who was the CEO of a company called Zynga back then, one of the big players in the space.

And he was giving a speech to junior entrepreneurs, to people that were trying to follow in his footsteps, and he said, "Do whatever you have to; to get as many users as you possibly can as quickly as possible. The other stuff will, you know, all kind of work itself out. Don't worry about privacy policies.

Don't worry about people's personally identifiable information. Get users. Just do that. That has to be your focus." And I worry that that's also the focus of a lot of A.I. companies. And again, not the big players, but a lot of these smaller A.I. companies that are trying to get that market share, trying to make something of themselves.

And so, I think that's why it's so important that government step in and that government provide a responsible regulation that doesn't stifle technology, but that does protect democracy and protect society. And I think that's exactly what we're talking about today. And somebody eluded to a little bit ago how it is a dynamic situation.

And when I first introduced a bill on deepfake technology back in 2019, something that I came to say over and over again is, you have the right to say what you want.

Minor exceptions apply, but the First Amendment being very strong in California, very strong in the United States, you don't have the right to put your words in my mouth.

And that's what this technology does, allows people to do, is put their words, their thoughts, their ideas into other people's mouths to give the misrepresentation that government officials or elections officials are saying things that they didn't say or do. And so that is what a lot of our legislative efforts this year are all about.

That was what that effort five years ago was about. And, you know, it's going to be, we've got a couple months left in the legislative process. There's still tweaking that we're doing, but we can't afford to not do not get this right in California and hopefully have that apply to the rest of the country.

Speaking of applying to the rest of the country or the rest of the world, and this is, I think, a question from Mister Harris, but you've talked about or yeah, I think for Mister Harris because it's more social media company related.

But are the social media companies that are having to change their policies and change their processes in the EU, are they also applying those changes globally, or are they just sticking to users in the EU and having more lax sort of policies and systems for other users and other viewers across the rest of the world?

That's a great question. Thank you for that, Assembly Member Berman, and thank you again for all of your work on this issue for so many years.

So, there are some things that social media companies are having to do to comply with the EU's laws that protect elections or that protect social media users more broadly that are having side benefits to the United States.

One really clear example was made apparent to me when, after giving a speech at Berkeley a few months ago, a very nice gentleman from TikTok came up to me after the speech and handed me his business card and a flyer advertising TikTok's researcher access programs.

So that letting me know as an academic that I could get early access to these programs. Now, TikTok is legally required to give researchers access to their data in ways that allow researchers to study things like elections or mental health in Europe.

They're not legally required to do that in the US, but they decided to go ahead and do that in the US largely. My guess would be a, because it's not a lot of additional marginal cost.

Once you build these tools for researchers to have access in Europe, you can just add in researchers in other parts of the world. And also, because, you know, as we all know, TikTok is under a lot of pressure right now and is trying to do anything they can think of to look like they're being a responsible player.

I think in other ways, though, we don't really fully benefit from the things that are happening in Europe because sometimes they're very specific, like the investigations that are taking place right now about the doppelganger campaign that I mentioned in Europe, like those investigations are being run by EU staff.

They require teams of people who work at the European Commission to enforce them.

So, we're not going to necessarily benefit from that unless by doing the investigations, it forces some of the social media companies to conduct investigations and then realize that these are actually global operations or realize that the accounts that are doing this in Europe are also accounts doing this in the US.

So, I think there's a mix, but I think, unfortunately, we can count on Europe as a source of inspiration, and we can occasionally expect to get some side benefits, that companies will be responsible in the ways that they're being legally compelled to be in Europe voluntarily in the US.

But I don't think we're going to get that from every company, even most of the time. And so, I think the better way to think about their laws is that the companies have proven that they can comply with them, or they've demonstrated that they're not complying with them in some cases.

But these laws have been tested in Europe, and if the companies can deal with it there, they can deal with it here, they can deal with versions of them here.

So I think better to think of us as having an opportunity to learn from what's working and what's not in Europe and then, you know, leapfrog and do an even better version of it here in California.

For sure. I appreciate that. And then last question. Bring it stateside. And I think the report references that there are a lot of states that have passed legislation in this space. Are there any states that we should be looking to that have passed legislation?

And full disclosure, this is a question a reporter asked me that I didn't have an answer to. So hopefully some of you all who know have been following this more closely. Are there other states that have been successful at passing legislation that you think we should be looking at to emulate or modify?

Yeah. So, you know, it's possible that Liz Howard from the Brennan Center will also have answers on this. I happen to have a list in front of me of existing statutes that are similar to both your Bill, AB 2655 and Chair Pellerin's bill, AB 2839. And those come from Michigan, Minnesota, Texas and Washington.

They're each distinct and have different pros and cons. But I think that they make some headway. In the space of A.I., there's a bill that was signed by the Governor of Colorado recently, just a couple of weeks ago. That's actually not election specific, but it is interesting in that it does create some accountability measures for A.I. companies.

And in some ways, it's more similar to a bill that just passed the Assembly by Assembly Member Bauer-Kahan: AB 2930 I believe so. The last count I heard was that there are over 400 pieces of state legislation relating to AI right now in the US that have been proposed.

So, I think in a couple of months we'll see what passes and what doesn't. There's a proposed piece of legislation in Vermont that I find very interesting. That's actually about liability standards for A.I. I think that it's really important for us to think about liability.

Vermont may jump ahead of Europe on this one, actually, because Europe is working on something called the EU A.I. Liability Directive that will take about three years before it's complete. But if Vermont is able to pass this A.I. law that they're working on, the liability standards could, I think, have very significant impacts.

Awesome. And I love Michigan, Minnesota, Texas, Washington. You've got blue states, you've got red states, you've got purple states. This is not a partisan issue. And you know, this is really an issue that just impacts every voter, no matter who you are or what your beliefs are. So that's great. We'll take a look at those. Thank you.

My pleasure.

Thank you. Assemblymember Essayli.

Thank you, Madam Chair. I have some questions. I'll start with Mister Harris because he brought up some interesting points and I want to delve a little deeper. I also want to say I think the focus here is supposed to be about A.I. regulation. This sort of seems to have drifted into disinformation regulation.

So Mister Harris, first I'd like to know what is your definition of disinformation? Is he up on the screen there he is. What do you consider disinformation?

Thanks so much for the question. And I appreciate the point that, you know, we're talking about these things and they're connected. Disinformation, you know, the, the definition that I use for myself is that it's information that's deliberately being distributed to mislead people.

Okay. What is misinformation?

So, misinformation is information that may be false, but people may not be deliberately distributing it. The same, the same piece of information could potentially be classified as both, depending on the intent behind the people distributing it.

And so, what I hear in your testimony is that you're concerned that social media isn't doing enough to regulate which one, misinformation or disinformation?

Well, actually both but I think that in this hearing and in this particular context, disinformation is of greater concern to us.

Okay, and so you think that it's private social media company's job to employ moderators to review posts and decide what's true, what's not true, and to take them down?

Well, no, actually. When I worked at Facebook, I was briefly part of a team that supported third party fact checking. And third-party fact checkers are usually associated with news organizations or actually are journalists. Reuters has a program that does this. The Associated Press has one that does this.

And so, the fact checking programs historically that are run by social media companies are in partnership with those third-party fact checkers. And they're not really any of the social media companies that want to be in the business of themselves operating as arbiters of truth. And that's why third-party fact checking is such an important tool.

It allows the social media companies to find information that may be circulating on their platforms that's been classified as by trustworthy third-party fact checkers, who, by the way, get credentialed by the International Fact Checking Network run by the Poynter Institute in Florida.

And then the social media companies have the opportunity to simply place a label that summarizes the information from the independent fact checker and a link that takes you to the website of that fact checker if you want to learn more.

I just...I'm baffled that we think that there could be some independent third party that we designate as the tellers of facts. So, you're saying that news companies, I so disagree with this model, news companies are the arbiters of the truth, which they've never gotten anything wrong.

And so, if they decide something's misinformation or disinformation, the social media should take that down. Let me ask you this. When they said in the 2020 election that the Hunter Biden laptop was Russian disinformation, was that true or not true?

So, I'm not an expert on that particular case, but I will tell you that the wonderful thing about third party fact checkers is that they're not necessarily partisan. And there are many different third-party fact checkers out there. And organizations from that don't have any political valence at all have become third party fact checkers.

And there are also news organizations from both the left and the right that are traditionally associated with the left and the right, who have also gotten into the business of third-party fact checking.

And if multiple fact checkers disagree, then that can be displayed in social media platforms as well, I'm personally not ready yet to give up on the entire institution of journalism or attempts to, to know what is true and what is false.

And that's why I believe that it's a good idea to have professionals who make it their full-time jobs, whether they're on the right or the left, to uncover the truth and report it to us and share it with us on many different types of platforms.

Okay, well, let me ask you this, because I wanted to get into this because we're talking about misinformation as it applies to elections. You said Russia is buying ads in Europe to influence the election, which you said is deja vu to what happened here. What are you referring to as the deja vu?

Well, in the lead up to the 2016 election, which is well documented in the Mueller report, there were two different operations financed by the Russian government. One was run by the Russian Internet research Agency, which was a business operated by the late Evgeny Prigozhin, a Russian oligarch. He ran it out of St. Petersburg, had an office where reportedly around 1000 people went to work every day to go and create fake content and interfere in elections. And in parallel, there was an operation run by the Russian GRU Military Intelligence Unit.

And both of those operations were heavily engaged in both producing large amounts of organic content, meaning content that they simply posted posing to be real people in the United States or posing to be organizations in the United States, but also purchasing ads, buying ads on those platforms, distributed.

I apologize to cut you off just in the interest of time. Did any of that have an impact on the election?

Well, the interesting thing is that the platforms that control the information have not historically made that information available to researchers who could answer that question and have chosen not to invest the funds or the time by researchers who work in those platforms to actually analyze what did happen in 2016.

So, unfortunately, that research simply has not been done, and people who have access to that information have chosen not to do it. And people who would like to do that research cannot do it.

There's no evidence anything that you just described had an impact on the election. Let me just ask this last question for you, and then I'll move on.

Okay. But Assembly Member Essayli, I just want to make sure we have time for other people and for public comment.

I understand.

Okay. Great. Thank you so much.

This Digital Services Act deal that you're suggesting that we implement, I don't understand. So how would a social company know what's a risk to democracy that's such a generic and broad term that encompasses basically everything. So how do you define what a risk to democracy is that's so broad?

Well, yeah, social media companies are big companies that have lots of different responsibilities and conversations that take place on them. Risks to democracy can take many different forms. A risk to democracy might mean someone spreading information that an election is taking place on a different day than it is actually taking place.

That might give people incorrect information about where to vote, how to vote. It could also mean people spreading information that is a deep fake. That has, for example, a fake picture of Donald Trump hugging Anthony Fauci, such as the one that was created by the DeSantis campaign.

So those are examples of risks to elections, where people can interfere in both the mechanisms of conducting the elections themselves and also where people can sabotage individual candidates through the use of distributing fake information.

Okay, I just. I'll save my closing comments for the end. I just have a couple questions for Miss Howard. Miss Howard, I assume you are a lawyer.

Miss Howard actually had to leave.

Oh, she left. I'm sorry. We have no lawyers. Okay.

But we have two other people, Assembly Member, who would like to speak. We need public comments, and then we could go to your closing.

Okay, we'll do that. I just - for the lawyer part, I'll just say, I'll close. Save that for closing. I did have one question for Miss Becky Waite. Do we still have her?

Yes, I am here.

Okay, Miss Waite, I see we put a lot of emphasis on trying to label things that may be A.I. generated or not true. And I often wonder, is it, do you think, a better approach or maybe something to look at when we want to label things that are authentic?

And so, I know a lot of times when I'm considering information, a post or something, I look at the source. I like to see, is this an authentic source? Is this legit? So, is that something you've given thought of?

Because I feel like the amount of incorrect or not authentic information is unlimited, but the amount of authentic, real, credible information is more limited. So maybe we focus on labeling those things.

Yeah, I think that's a great question. And I think Providence is - I talked a little bit about Providence in my opening remarks. I think the issue of Providence is really two verticals.

One is helping people understand the origin of a piece of content, regardless of where the origin is, if it's quote unquote authentic or generated or modified or altered in some other way. And the work that we're doing with C2PI, I think is best suited for that. A little bit of context on how that works.

C2PI is a piece of metadata that's attached to a piece of content. You can think of it a little bit like a passport, so it travels around with an image or a file and allows you to see where it originally came from. The thing that's great about C2PI is it's really hard to fake.

You can't pretend it came from somewhere it didn't, but it's easy to remove. And I think exactly to your point, that lends itself very well to labeling potentially authentic images. In fact, C2PI is not just a coalition of tech industry bodies, but also includes other industries, including the media camera makers.

And I think that is likely the future of where Providence is going. There are other tools that are better for identifying something that has been generated or altered, and I think that's a subset of a different problem. But I'm very excited about the sort of future work on labeling authentic content.

And generally, just so I understand, if I'm online, how would I know something is authentic? Is it going to have like a label or do we use a third party authenticator? What does that future look like?

Yeah, so C2PI, there's a publicly available website that you can go on and drop in a file. And if that file has C2PI on it, that Metadata is attached. It will tell you where it came from, the origin of that piece of content.

So, I'm sure as this technology gets better, we could have things sitting on our phones to tell us what's authentic or what's not.

You know, I think a great example here is an image by a journalist. You want to be able to verify that that is authentic. It came directly from a camera. It hasn't been altered or modified in some way. And C2PI allows for that.

That's great. Thank you.

So, yeah, Providence is a complicated issue. You need a lot of different ways to address it, but agree with you that authenticity is a major, major area.

I appreciate it. Thank you.

Okay, great. Assembly Member Weber.

Thank you, Chair, and thank you to both Chairs for actually putting this very important informational hearing together.

And for all who came to sit on the panel, I think at least the majority of us are aware of the fact that A.I., this new emerging technology, is what is being used to create both misinformation and disinformation, which has a significant impact on our upcoming elections, which is why those two terms have come up so much in this hearing.

But to kind of piggyback off of what was just stated in the area of Providence and focusing on authentically generated data. I've seen those watermarks on pictures and you just kind of click on it and it tells you all of the data.

And we're working at least with the bill that I have to ensure that social media companies can't stress strip the Providence data.

But wondering from the Secretary of State's Office, since there's a lot of images that you all put out there to ensure the public that this is actually something that is authentic, have you all looked into using any of these kind of watermarks or anything to put in your images so that people can go and tell that this is an authentic generated image versus something that may not be authentic?

I'm going to turn it over to our communications director.

This is new and we are looking into it. Yes, thank you.

I would definitely, strongly recommend that you all, as was just stated, C2PI. This is something that is publicly available.

I believe it's also for free, but it is game changing for people to be able to go and look and kind of click on something if you actually have that watermark image to determine that this is something that is authentically generated.

But definitely want to thank you all for being here and thank the Secretary of State's Office and all of the staff for all of the work that you all are doing to ensure that the myths and disinformation campaigns will have as minimal impact on our elections as possible here in California, and that you are proactively getting out there to help out with all 58 of our counties for this very important upcoming election.

So thank you.

Okay, thank you. We would like to invite any members of the public who would like to give public testimony to come forward to the microphone. Is anybody here in the room wanting to offer any public testimony? Okay. Seeing none, I won't have to Institute time limits on those public speakers.

So I want to say thank you to everybody who came today. Thank you to the Secretary of State's Office who took off the first row, many of you, and participated and watched closely the whole hearing. I appreciate that.

And to our Zoom participants as well, who took time out of their day to come and be here with us. I also want to thank the Elections Committee on the assembly side and the members of the committee for taking time out of your busy schedules to come.

If you were not able to testify today, please submit your comments to the Senate Elections Committee and Constitutional Amendments, or to the Assembly Committee on Elections. Your comments are important to us, and we want to include your testimony in the official hearing records.

I'd like to make sure that my colleagues who would like to make closing statements are able to do so. Would anybody like to do that? Yes. Okay. Assemblymember Essayli.

Thank you, Madam Chair. I appreciate the opportunity. I do think this is a really important conversation and we should look at specifically what concerns A.I. raise. Now, many of the things that were brought up here, Madam Chair, I would say a lot of what was talked about is already illegal. It is illegal to intimidate voters.

It is illegal to impersonate an election official or a government official. So maybe our conversations or legislation needs to look at giving law enforcement tools to better go after that, or increasing penalties and punishment for people who do those things.

I will say one concern I do have from today's conversation is what I'm hearing and what I'm feeling, especially from Mister Harris, is this idea that there should be an arbiter of truth that either the government requires, or the government creates this entity that becomes the arbiter of truth in our society.

And I do have serious concerns about that. The Supreme Court has recognized that false information or false speech is protected speech. There has to be breathing room for open debate and discourse that may include false ideas. And so, the Supreme Court is very reluctant to uphold laws that appear to regulate what's true or what's not true.

It's one of the issues I also have with these moderation employees that Mister Harris thinks we should have more of in Twitter, where it was known that the FBI and the government was going to these social media companies and telling them to take down people's posts. It was government censorship.

They were censoring about COVID information, they were censoring about election information. They took down all the information about Hunter Biden's laptop, claiming it was false information. We now know it wasn't. So, this is the problem when we have people deciding what the public can and can't know.

So, I would really suggest that we focus our intent and how we protect the public from actual illegal conduct and not get into the truth regulating business, that's my biggest takeaway from today. And I hope that we can make it bipartisan and work together on improving this.

And I do think the future is, like my colleague said, authenticating real information and making sure people know what's real. And I do think there's a huge appetite for that. The public wants, they want authentic information, and they want to know they could easily decipher it from fake information. With that, I yield back. Thank you okay.

Thank you. Chair Pellerin.

So again, I just want to thank our Chair, Blakespear, for putting on this Joint Hearing.

Thanks to all the speakers who are here, and thank you to all those elections officials out there who are doing the incredible work that needs to be done to conduct a fair, safe, accurate, transparent election for everyone and elections that are accessible to everyone and ensuring that the information that's disseminated is correct.

So, people can make informed decisions, is absolutely imperative for our democracy. So, thank you all for being here, and thank you for holding this hearing.

Sorry, just last comments. Thank you again for everybody and for the staff of the committees for putting this on.

Okay, so we're near adjournment. Oh, Assembly Member Berman. Go ahead.

I didn't have Hunter Biden's laptop sneaking into the Committee hearing, but I would note that his trial today shows the nonpartisan nature of our courts and how much we should respect the decisions that our courts and the juries of people's peers make in our courts in America and what a beautiful foundational element of this country that is.

But also, just really want to note the dangers that are being created by this new technology that's constantly coming out, constantly being improved, constantly being democratized and making it easier for people to utilize, many times for good purposes, but sometimes for bad purposes as well, and how we don't want social media companies to necessarily be the deciders of what's a lie and what's the truth.

But we can create regulations that say that when there's manipulated content, when there's manipulated media, when there are manipulated images that create the impression of somebody saying something that they did not do or say that that is something that can easily be regulated and easily be one of the many things that we do to protect our democracy here in California and hopefully have that impact other states as well.

So I think that's what we're trying to accomplish here. I think it's a righteous, patriotic effort, and I'm glad that we're having those conversations. Thanks.

Thank you. Well, thank you, everybody. And with that, we are adjourned.