Lawmakers hear expert testimony on S.23, bill on synthetic media in elections

Representatives on the House Government Operations & Military Affairs committee heard expert testimony April 16 on S.23, an act relating to the use of synthetic media in elections, as witnesses described state approaches to regulating deepfakes and raised First Amendment concerns.

The National Conference of State Legislatures’ Adam Cook Hook and Wendy Underhill told the committee that 23 states have enacted laws addressing synthetic media in elections and that the laws vary widely in scope and enforcement. "When we are referring to deepfakes today, we are talking about computer programs that create images, audio or videos," said Adam Cook Hook, a member of NCSL’s Elections and Redistricting team. He noted examples of tools such as MidJourney, Sora and Adobe Firefly.

Why this matters: committee members said they want a clearer picture of what S.23 would require and how enforcement would work if the state adopts disclosure or prohibition rules. Lawmakers and witnesses repeatedly flagged tensions between efforts to prevent deception and the constitutional protections that govern political speech.

Hook summarized how states have divided approaches: two states (Minnesota and Texas) prohibit publication of an election-related deepfake a set number of days before an election — 90 days in Minnesota and 30 days in Texas — while most other states adopt disclosure requirements similar to campaign-finance labeling. "The other 21 states with laws on this subject have some type of disclosure requirement," he said. He also described less-common measures such as digitally embedded metadata disclosures (used in Colorado and Utah) and platform-moderation requirements in California.

NCSL presenters outlined enforcement mechanisms used in state laws: civil suits seeking injunctive relief, civil fines (New Mexico and Utah were cited as $1,000-per-violation examples), and, in about one-third of states, criminal penalties ranging from fines to short jail terms. Hook cited Coles v. Bonta, a California case in which a court granted a preliminary injunction against a state law restricting parody deepfakes while the case proceeds.

Wendy Underhill, director of NCSL’s Redistricting and Elections Team, said most practical use of deceptive synthetic media in U.S. campaigns remained limited through the 2024 cycle, but that technology and the legislative response had moved quickly. "We are going to talk about AI and how it relates to political messaging," Underhill told the committee. "We’re looking at places where it could be used for deception." She and Hook also provided links to public databases that track examples of synthetic media worldwide.

The ACLU of Vermont urged caution in drafting broad new mandates. "S.23 specifically does put burdens on speech about candidates and raises constitutional concerns," said Falco Schilling, advocacy director for the ACLU of Vermont. Schilling told the committee that compelled disclosures and content-based restrictions on political speech draw strict scrutiny under the First Amendment and that broad labeling mandates that apply outside paid mass media are legally risky.

Schilling noted several areas of particular concern in S.23 as discussed at the hearing: vague statutory definitions that could sweep in ordinary image edits (Photoshop, Lightroom), narrow or difficult-to-apply exceptions for satire and parody, and private civil suits that could be used to obtain injunctions stopping speech during an active campaign. He suggested narrower alternatives, such as focused prohibitions on misinformation that misrepresents time, place or legal consequences of voting, or strengthened source disclosures tied to paid communications.

Committee members asked technical and enforcement questions. Representative Hooper asked whether metadata embedded in a file would be accessible to platforms and users; Hook said embedded metadata is generally accessible and that platforms could, in principle, read it on upload. Representative Hango and others asked about the practical effect of disclaimers on public trust; Underhill and Schilling cited a study from the NYU Center on Tech Policy finding that labeling edited or AI-generated content can reduce audience trust in the speaker and that the wording of a label matters.

No committee vote was taken on S.23 during the hearing; the session consisted of witness presentations and member questions. Committee members asked staff to continue gathering materials and offered NCSL and the ACLU the opportunity to provide follow-up information.

Ending: With lawmakers still weighing legal and practical trade-offs, the hearing closed with a request for additional research and for witnesses to supply written resources that the committee could use while drafting or amending the bill.