AI and Elections: The State Disclosure-Law Landscape

No state has enacted an outright federal-level ban on AI-generated content in political ads — the pattern that has actually reached the statute books is a disclosure or labeling requirement: political communications that use AI-generated or synthetic media must say so, usually within a set window before an election. At least 16 states have enacted 20 separate statutes doing some version of that since 2024, Congress has introduced more than 150 AI bills without enacting one, and five closely watched 2025-session state bills — in Alaska, Arkansas, California, Connecticut, and Georgia — remain unresolved. This guide is a snapshot of that landscape as legislative tracking organizations had it dated between July 2025 and June 2026, including the specific category CASRAI’s own NIKOLAI project catalogs this risk under in frontier labs’ safety frameworks.

What state “AI election” laws actually require

Every bill covered here follows the same basic shape, whether enacted or still pending: it does not ban synthetic media in campaign communications outright, it requires that AI-generated or manipulated audio, video, or images be labeled when used in political advertising or electioneering communications, typically inside a defined pre-election window (Connecticut’s failed 2025 bill specified 90 days). A smaller number of bills work the opposite direction — California’s A502 would exempt an election communication from an existing AI-content prohibition specifically because it carries an AI-generated disclosure statement, treating disclosure as a safe harbor rather than a standalone mandate.

• Common trigger: use of AI-generated, synthetic, or manipulated media depicting a candidate or event in a political ad or electioneering communication.
• Common remedy: a disclosure statement on the communication itself, not a takedown or ban.
• Common window: a defined number of days before an election (where specified).
• Common exclusion: parody, satire, and news-reporting exemptions appear in most of these bills’ text as tracked by NCSL, though the exact carve-out language differs by state.

Five bills NCSL was tracking as of mid-2025

NCSL’s “Artificial Intelligence 2025 Legislation” summary (last updated July 10, 2025) singled out five state bills specifically addressing deceptive synthetic media in elections:

“Carryover” is a legislative status, not a prediction: it means the bill survived its chamber’s session-end deadline and remains eligible for a floor vote when the legislature reconvenes, without having to be reintroduced from scratch. It says nothing about whether it will ultimately pass.

The states that already have a law on the books

While those five bills were moving through 2025 sessions, other states had already finished the job. NCSL’s cross-session “AI in Elections and Campaigns” tracking table (last updated June 23, 2026) lists 20 enacted state statutes across 16 states, dating from the 2024 and 2025 sessions:

NCSL classifies all 20 under its “Deepfake” topic tag, which it defines as legislation addressing “deceptive synthetic or manipulated audio, video, or images that realistically depict individuals or events” — the same disclosure-first model described above, not a content ban.

Status shifts by mid-2026 — and why that matters for a page like this

Checking NCSL’s two tracking pages against each other, dated eleven months apart, surfaces a real example of why a legislative snapshot needs a visible timestamp. The July 2025 page listed Connecticut’s H 6846 as failed. The June 2026 table lists the same bill as pending — consistent with a bill that was reintroduced, or whose earlier “failed” reflected only that chamber’s session rather than the end of the two-year legislative cycle. Arkansas’s H 1041 and H 1141 are listed as failed in both. Alaska’s three bills and Georgia’s S 9 remain “pending – carryover” in both. No additional 2026-session enactment specific to election deepfakes appeared in the sources checked for this page as of the June 2026 table’s own last-updated date; the 20-state, 16-bill enacted count above is not a 2026-session count, and readers relying on this page for a current whip count should re-check NCSL’s live tracker rather than treat any single dated snapshot — including this one — as still current.

Congress: 150-plus bills, zero enacted

The federal picture is far less active than the state one. The Brennan Center’s Artificial Intelligence Legislation Tracker (last updated September 26, 2025) counted more than 150 AI-related bills introduced in the 118th Congress addressing restrictions on AI systems, evaluation requirements, transparency mandates, new regulatory authorities, consumer protection, government studies, and AI procurement policy. None became law. The 119th Congress has introduced new and reintroduced proposals, but as of that tracker’s last update, no federal statute specifically addresses AI-generated content in elections. Election-AI regulation in the United States is, so far, an exclusively state-level activity.

How this differs from California’s AI Transparency Act vs. SB 53

CASRAI’s existing comparison of California’s AI Transparency Act and SB 53 covers two different California statutes: SB 942/AB 853, which sets general consumer-facing AI-content labeling requirements, and SB 53 (the Transparency in Frontier Artificial Intelligence Act), which requires the largest AI developers to report catastrophic-risk safety practices. Neither one is election-specific, and neither appears anywhere in that comparison’s text. California’s A 502, discussed above, is a third, separate California statute working on a different axis entirely — it doesn’t set a general labeling duty or a frontier-model reporting duty, it carves an exemption into an existing election-communications restriction when an AI-disclosure statement is present. This guide is the first page in CASRAI’s frontier-AI-safety content covering that election-specific axis; readers who came here from the SB 942/SB 53 comparison looking for California’s elections rule specifically should look to A 502, not either statute in that comparison.

How CASRAI’s NIKOLAI tracks this

State election-deepfake statutes regulate a real-world instance of a risk category CASRAI’s own NIKOLAI project catalogs across frontier AI labs’ published safety frameworks. NIKOLAI is CASRAI’s own independent, unendorsed reference dictionary of frontier-AI-safety terminology — not an official standard, and nothing here has been declared or confirmed by any lab, legislature, or regulator. NIKOLAI’s Actors, models and scope track (N1) defines a risk domain element as “a proposed top-level category of catastrophic harm…under which threat models and thresholds are grouped.” Reading published frontier-lab safety frameworks side by side, that element identifies harmful manipulation — “strategic distortion of human behaviour or beliefs by targeting large populations or high-stakes decision-makers through persuasion, deception, or personalised targeting” — as one of the four categories that recur most often, alongside CBRN, cyber offense, and loss of control.

The labs themselves are inconsistent about it, which CASRAI’s companion guide on why “AI R&D” and harmful manipulation live in four different places documents in more depth: OpenAI’s Preparedness Framework explicitly routes persuasion risk outside its own Tracked Categories, Anthropic’s Responsible Scaling Policy doesn’t name harmful manipulation among its own threat models at all, and xAI’s Frontier AI Framework names it as one of “four primary risk domains” in its introduction and then never writes the corresponding “addressing” subsection for it. Election-targeted synthetic media is a specific, real-world case of exactly that unsettled abstract category — which is part of why state legislatures, not the frontier labs’ own frameworks, are where the most concrete rules against it currently sit. None of the state bills above cites NIKOLAI, and no legislature or lab has filed a NIKOLAI Mapping Declaration connecting election law to this element; the link drawn here is CASRAI’s own reading, a shadow mapping only.

Frequently asked questions

Have any states actually enacted a law on AI deepfakes in elections?

Yes. NCSL’s cross-session tracking table (last updated June 23, 2026) lists 20 enacted statutes across 16 states since 2024, including Alabama, Arizona, Colorado, Florida, Idaho, Indiana, Mississippi, New Hampshire, New Mexico, New York, Oregon, Kentucky, Montana, North Dakota, South Dakota, and Tennessee.

What do these laws typically require?

A disclosure or labeling statement on political communications that use AI-generated or synthetic media, usually within a set window before an election — not a ban on using AI in campaign content.

Is there a federal law on AI-generated election content?

No. The Brennan Center’s federal tracker counted more than 150 AI bills introduced in the 118th Congress and none enacted, as of its September 26, 2025 update. No federal statute specifically addresses AI in elections.

What happened to Alaska’s, Arkansas’s, California’s, Connecticut’s, and Georgia’s 2025 bills?

As of NCSL’s two most recent tracking pages (July 2025 and June 2026), Alaska’s S 2, S 33, and S 64 and Georgia’s S 9 remain pending-carryover, Arkansas’s H 1041 and H 1141 failed, California’s A 502 is pending, and Connecticut’s H 6846 — listed as failed in July 2025 — is listed as pending in the June 2026 update.

Does this overlap with California’s SB 53 or the AI Transparency Act?

No. Those cover general consumer AI-content labeling and frontier-model catastrophic-risk reporting, respectively — see CASRAI’s comparison of the two. California’s election-specific AI bill is the separate A 502.

Is this page current?

It reflects legislative tracking dated between July 2025 and June 2026, sourced from NCSL and the Brennan Center. State legislative status changes continuously; check NCSL’s live tracker before relying on any bill’s status here as current.

Related reading

• C2PA Content Credentials: How Provenance Actually Works
• California SB 243 (Companion Chatbots): The State’s Second AI Reporting Channel
• State AI-in-Therapy Laws Compared: Illinois WOPR Act vs. Nevada AB 406 vs. Utah HB 452
• Italy’s Law 132/2025: A National AI Statute Inside the EU AI Act
• AI Claim Denials: The State Laws That Require a Licensed Human to Make the Medical-Necessity Call