Four instruments, the dates they carry, and — for each one — what it asks of you and what it doesn't. Every fact is checked against the document and linked. Where the text isn't published, the page says so instead of guessing.
Last verified October 26, 2026. Corrections: [email protected]. We're not lawyers and this isn't legal advice.
The FEC's AI-in-campaign-ads rule has been in force since
United States (FEC)
The only operative federal rule on AI in campaign ads is the technology-neutral fraud ban at 52 U.S.C. § 30124. The FEC declined AI-specific rulemaking after considering more than 2,000 comments, and its majority said the agency lacks legislative authority for anything broader — while asking Congress to widen the ban itself.
Nothing to do. There is no AI-specific safe harbor: whether an ad violates the ban is decided case by case.
California requires an AI disclosure on election ads
California
Gov. Code § 84514: committees paying for election ads must carry the exact statement "Ad generated or substantially altered using artificial intelligence." Regulations adopted August 2026 add a three-part test for "materially altered" (cartoons excluded) and require committees to retain the AI source or reference media with its metadata — that regulation's effective date was not confirmed in the documents we reviewed.
If you pay for election ads in California, the disclosure is mandatory and the wording is fixed. Keep the source media and metadata.
Colorado's ADMT and chatbot rules close for written comment
Colorado
Comments close at 11:59 p.m. as printed on the docket. Colorado is on daylight time until November 1, so that reading is an hour off — if you are close to the deadline, treat 10:59 p.m. Pacific as the real one. What it decides: whether the interim draft's "materially influence" presumption survives. As drafted, a decision is presumed AI-influenced whenever a human sees an AI output and acts consistently with it, rebuttable only by showing the output was de minimis. That clause sets which everyday automated decisions owe notice, explanation, correction, and human review.
This is the last public input before the record closes. The comment we would send, to edit into your own:
Please revise the "materially influence" presumption so a decision is not treated as AI-influenced whenever a human sees an AI output and acts consistently with it. As drafted the only rebuttal is showing the output was de minimis, which will pull in decisions the system barely touched. Please state in the rules which decisions are covered, with dated examples, before they take effect January 1, 2027.
White House draft language defining "Super Intelligence" is due
United States (White House)
Sixty days from the September 29, 2026 executive order that renamed AI across the executive branch. Internal executive output — a watch item, not a public process.
Nothing to do. It is not open for comment and does not reach you directly.
Pennsylvania's deceptive AI advertising act takes effect
Pennsylvania
Act 99 of 2026 reaches advertisers of goods or services who use synthetic or significantly AI-modified image, video, or audio that creates a false consumer perception without clear and conspicuous disclosure. Text-only ads, evident fiction, and parody or satire are excluded. Enforcement runs through the Attorney General and district attorneys: injunction plus civil fines up to $3,000 plus fees. There is no private right of action, and "advertiser" is left undefined — agency interpretation is the live question.
If you advertise in Pennsylvania, the disclosure is the safe harbor, and it has six criteria to meet: size, contrast and duration, adjacency, non-contradiction, audio volume and cadence, and print prominence. It must stay up "to the extent technically feasible."
Illinois' AI Safety Measures Act takes effect
Illinois
Scoped to the largest AI developers, not small firms: independent third-party audits and 72-hour incident reporting. Enforcement is by state regulators.
Nothing, unless you are one of the largest developers. The threshold is what decides that, and we have not yet verified the statute's own link — treat the trade coverage as the source until we do.
Colorado's ADMT and chatbot safety rules take effect
Colorado
The rules implementing SB 26-189 and HB 26-1263. The docket requires final rules to take effect on this date.
If you run automated decisions that reach Colorado residents, this is the date the notice, explanation, correction, and human-review duties attach. The covered-decision list in the final rules is what decides whether that includes you.
The 119th Congress ends
United States
Every bill on this page that has not been enacted dies here and would have to be reintroduced. Both AI bills we track sit at the same point today: referred, no hearing scheduled — H.R. 10768 with zero cosponsors, H.R. 10814 with four. Cosponsors, a hearing, and a markup are what predict reintroduction.
Nothing to do, unless one of these bills moves — and if it does, this page changes first, with the date and the document.
New York RAISE Act New York
Seventy-two-hour incident reporting for large developers. An enforcement date of January 1, 2027 appears in secondary reporting only — verify against the chaptered statute before relying on it.
California SB 53 California
Frontier-lab disclosure duties. The effective date was not stated in the sources we reviewed.
H.R. 10841, Securing Elections From AI Deception Act, if enacted United States
Text published; not an FEC ad-disclosure bill. A criminal voting-deprivation prohibition effective 60 days after enactment, plus FTC disclaimer and duty-of-care regimes 60 days after later regulations — up to two years for the FTC rulemaking. Absent passage it dies January 3, 2027.
H.R. 10768
It would create an Office of Artificial Intelligence and Emerging Technology Threats inside the Executive Office of the President.
Introduced · Energy and Commerce · text unpublished
Not yet, and possibly never. This bill creates a federal office; it does not regulate anyone outside the government.
H.R. 10814
It would require political ads to state, inside the ad, when generative AI produced an image or video in them.
Introduced · House Administration · text unpublished
Only if you pay for federal political advertising. Ordinary readers: not at all.
H.R. 10841
Criminal penalties for AI used to deceive voters, plus FTC-enforced disclaimers
Introduced · Energy and Commerce
Developers and deployers of covered algorithms, in the narrow case of election deception. Most readers: no.
PA SB 806
Pennsylvania bans AI-generated deceptive advertising, enforced by the Attorney General
Law
If you advertise in Pennsylvania, in 56 days.
Each page carries the same six parts: where the bill is, what it does, what you should be aware of, what we don't know, what we think you should do, and what we're trying to accomplish. The first four are description and every claim in them is linked to a document. The fifth is our view, and it is labeled as ours. The sixth is the same on every page, because it's the reason the pages exist.
Three things follow from that. If a bill asks nothing of you, the page says so — a page that can't tell you to stand down can't be trusted when it tells you to act. If the text hasn't been published, the page says that too, and labels what came from the sponsor's own draft. And when a bill changes, the facts at the top update while the analysis stays where it was, with a note that it was written against an earlier version. A page that quietly rewords itself is worse than a stale one.
We watch who ends up holding the verification — the evaluator, the auditor, the office that decides what counts as safe. That's where the power lands in any rules regime, and it's the thing most coverage of these bills leaves out. Back to the AI index →
