Testimony of Paul Seamus Ryan, Executive Director of the New York City Campaign Finance Board to New York City Council Committee on Governmental Operations Committee on Proposed Int. No. 12-A: Requiring Disclosure of Digitally Manipulated Content in Local Elections
June 22, 2026
Good morning, Chair Brewer and members of the Committee on Governmental Operations, State & Federal Legislation. My name is Paul Seamus Ryan, and I am the Executive Director of the New York City Campaign Finance Board (CFB). Joining me today is First Deputy Executive Director Amanda Melillo. Thank you for the opportunity to testify today regarding Int.12-A of 2026.
The CFB is an independent, nonpartisan City agency dedicated to strengthening local democracy. The CFB combats the influence of big money in politics by amplifying the power of small-dollar contributions from everyday New Yorkers through our matching funds program. We work to eliminate barriers to participation by equipping New Yorkers with the information and resources they need to vote and run for office. Through our NYC Votes initiative, the CFB engages directly with voters and provides resources to make voting more accessible.
Int.12-A, sponsored by Speaker Julie Menin, would require disclosure to the Campaign Finance Board when candidates for local elected office, or entities making local-election-related independent expenditures, create political materials or advertisements using artificial intelligence software capable of generating synthetic content. The CFB previously provided testimony on an earlier version of this bill, Int. 293-2024, on December 4, 2024.
The CFB acknowledges the Council's efforts to protect the integrity of our local democratic process as artificial intelligence technologies advance. The CFB has a unique perspective on this topic rooted in our dual mandate to serve candidates and voters. We sympathize with the perspectives of both. We wholeheartedly share your concerns about the ways that AI technology could be used to spread false information to intentionally mislead or deceive voters about Council members and other City candidates and officeholders.
At the Campaign Finance Board, we have placed a priority on learning about artificial intelligence. Ms. Melillo and I recently completed the Harvard Kennedy School AI in Action certificate course, we have formed an agencywide AI working group, and we are learning more every day. Nevertheless, AI technology is evolving at such a rapid pace that we cannot say with certainty how best to regulate its use in elections.To that end, we appreciate that the amended version of this legislation meaningfully addresses our most significant legal and implementation concerns shared at the hearing in December 2024. By shifting from a prohibition to a disclosure model and naming the CFB as the agency in receipt of these disclosures, as well as naming the CFB as the agency responsible for implementing the new disclosure regime, the Council eliminated the greatest vulnerabilities to First Amendment legal challenge and clarified administrative and enforcement responsibilities.
The CFB is well qualified to administer and enforce the requirements of Int. 12-A. We receive disclosure reports from all campaigns running for city office and all independent spenders active in city elections, and we impose fines and penalties for violations of our campaign finance rules and laws. Importantly, this bill casts the CFB in a role we already play as a disclosure administrator—receiving, processing, and making public the information reported to us—not as a content regulator. For that reason, we are relatively confident in our ability to implement this new mandate, though significant resources and time will be required.
Shifting the strategy to a disclosure model means the Council would be placing an entirely new administrative and enforcement mandate onto the CFB. We want to be realistic about the timeline and resources that would be required to implement this legislation. We are still researching the exact technology and staffing costs associated with implementing the bill, but we anticipate that the cost would be steep and that it would take at minimum a year to implement after bill passage. This estimate includes both updating our existing candidate and independent spender reporting systems and building an entirely new internal system to receive and make public the provenance data and other similar ad information we would be receiving from candidates and independent spenders.
As noted moments ago, the legislation’s shift from prohibition to disclosure largely mitigates the First Amendment risks present in the original draft. The Supreme Court has consistently upheld disclosure requirements against constitutional challenges over the past half-century, including in Citizens United, McConnell v. FEC, and in the Court’s seminal 1976 campaign finance decision in Buckley v. Valeo. Over and over, the Supreme Court has recognized the governmental interest in providing the electorate with information necessary to make informed choices in the political marketplace. Int. 12-A serves this interest. To even further strengthen Int. 12-A, the Council should consider narrowing the law’s application in several ways.
Int. 12-A incorporates by reference a definition of “synthetic content creations system” in the state Stop Deepfakes Act. The most recent version “E” of that legislation, which was passed by the Assembly on June 4th, defines the term to mean “a class of generative artificial intelligence system capable of generating wholly synthetic content.” This is a very broad standard, encompassing not only ads that contain synthetic content, but also ads generated or modified using a system capable of generating synthetic content even if such synthetic content was not incorporated into the ad.
The earlier version of the Stop Deepfakes Act that was passed by the State Senate on June 3rd exempted from this definition “technologies such as red-eye filters or other technologies that are only capable of making changes to existing audio or visual content.” I will return to the significance of this legislative language in a moment.
Another aspect of Int. 12-A’s broad coverage is its application to “any literature, advertisement, or other communication.” Combined with the Stop Deepfakes Act’s definition of “synthetic content creations system,” Int. 12-A would seemingly require a campaign that uses generative text software like ChatGPT or Claude to draft or edit a fundraising email, for example, to report this communication to the CFB.
A third element of Int. 12-A’s breadth is that its candidate disclosure requirements are triggered when a candidate “makes, publishes, or publicly distributes” a covered communication. This language would seemingly encompass instances where a candidate or campaign staff member reposts someone else’s content on a social media platform like X, Instagram or TikTok—without spending any money and perhaps with no knowledge of whether the content was created or modified with a synthetic content creations system.
The Council should look to the federal AI Transparency in Elections Act (S. 3875), principally sponsored by Senator Amy Klobuchar, for an example of a narrower scope of regulatory coverage.
First, rather than covering all forms of communication, such as fundraising email created or edited with ChatGPT or Claude, the federal AI Transparency in Elections Act applies only to “an image, audio, or video that was substantially generated by artificial intelligence.”
Second, the federal bill’s coverage is narrowed further by exempting images, audio, and video that have only minor AI alterations, “including cosmetic adjustments, color editing, cropping, resizing, and other immaterial uses” that do “not create a fundamentally different understanding than a reasonable person would have from an unaltered version of the media.” This approach is similar to the version of the state Stop Deepfakes Act passed by the State Senate on June 3—an exemption that was removed from the version passed by the Assembly on June 4.
Finally, the Klobuchar bill applies only when a person makes a “disbursement for the purpose of financing a covered communication”—effectively exempting from the law candidate reposts of social media content. As an alternative or complement, the Council could apply the disclosure requirements of Int. 12-A only in instances where candidates know or should reasonably know that the media they are distributing contains AI-generated content.
Narrowing Int. 12-A in these ways might better serve the public interest. As currently written, Int. 12-A requires disclosure not only of malicious synthetic content but of all political communications that were, in whole or in part, created or modified using software that is simply capable of generating synthetic content. Given the ubiquitous incorporation of generative AI capabilities into design and other software, and the widespread use of ChatGPT, Claude and other text generation software, Int. 12-A could require disclosure of nearly all political communication.
The CFB may end up with a database filled with information about innocuous, immaterial uses of AI, muddying the water for journalists, law enforcement, and members of the public seeking information about malicious use of artificial intelligence for deepfake images, audio and video. We may inadvertently bury the needle of information the public wants and needs in a haystack of information that serves little or no public interest. Limiting the disclosure law’s coverage in the ways I suggested moments ago would help reduce the volume of extraneous information.
Even with the refinements we suggest, we want to be clear-eyed about the practical limitations of this regulatory approach, so as not to create unrealistic expectations. The combination of low cost, high quality deepfake technology and social media platforms enables bad actors to spread disinformation in elections with speed and effectiveness unfathomable just a year or two ago. Earlier this month, The New York Times ran a disheartening profile of the world’s leading deepfake expert, Professor Hany Farid, who has given up on his own ability to distinguish deepfakes from reality and moved from the Bay Area to a cabin in the Green Mountains of Vermont.
Unstated in the Times profile is that Professor Farid is an advocate for digital provenance systems of the sort reflected in Int. 12-A. However, the real utility of digital provenance systems is enabling people to trust verified images and video as true and to dismiss unverified images as untrustworthy. Digital provenance systems serve as a sort of digital media nutrition label. Candidates would likely comply with the disclosure requirements of Int. 12-A, and that’s a good thing.
Unfortunately, even with the enactment of Int. 12-A, the public will still have access to the equivalent of digital junk food created by bad actors who may be unlikely to comply with the nutrition label-like disclosure requirements of Int. 12-A. The increasing difficulty of distinguishing deepfakes from truthful media, combined with the anonymity offered by social media platforms and other Internet-based communication streams, may present enormous enforcement challenges for the CFB.
Nonetheless, we remain deeply committed to working collaboratively with the Council to ensure this becomes an effective piece of legislation. While we have outlined specific concerns today, our ultimate goal is to partner with you to provide realistic guidance on resources and funding needed, refine and improve the legislation, and deliver a bill that truly serves New Yorkers.
The CFB is committed to ensuring our City's elections are transparent and secure. Thank you for your time and your leadership on this issue. Amanda and I are available to answer any questions you may have.