The AI Rules That Actually Passed This Year (and the Ones That Didn't)
A plain list of which AI laws are real as of September 2026, which are still proposals, and what you owe either way
By John Martines · September 27, 2026 · 7 min
Kill switches, superintelligence bans, auditor registries: September was loud. Here is what actually became law by September 2026, with dates and places, what is still only proposed, and the older rules that already cover AI in your business.
This isn't a month where the rules settled down. It's a month where the headlines got louder than the laws.
In September alone you could read about a California kill switch, New York City kill switches, and a federal bill to ban superintelligence outright. Some of that is law. Most of it isn't. If you run a 20-person company, the difference matters — one changes what you do in January, the other is something to watch over coffee.
Here's the sort, as of September 27, 2026. This is a plain summary, not legal advice. If one of these touches your business, have your attorney read the actual text.
California, SB 1050: AI performers in ads must be labeled. Governor Newsom signed it on September 16, 2026. It takes effect January 1, 2027. If a video or audio-only ad uses a "synthetic performer" — an AI-generated human-like figure or voice that isn't a specific real person — the ad needs a clear, conspicuous disclosure close to that performer, in wording substantially similar to "this performance features a synthetic digital performer." A violation counts as false advertising under California law, and a noncompliant ad has to be pulled or fixed, not left to run out its flight.
This one reaches small businesses directly. If you've started making promo videos or radio spots with an AI voice or an AI spokesperson and they run to California customers, you're in scope.
California, SB 813 and AB 1405: a system for AI auditors. Signed September 9, 2026. SB 813 has the state set qualification rules for "independent verification organizations" that test AI safety claims, with the state's deadline in 2028. AB 1405 creates a state registry of AI auditors with independence and ethics standards, with the registry due by 2029. As written, this is voluntary infrastructure — it builds the auditing profession, it doesn't order your company to get audited. Where you'll feel it is later, in vendor contracts and questionnaires that start asking whether a tool has been independently checked.
Colorado, SB 189: the AI Act, rewritten. Governor Polis signed it on May 14, 2026. It repeals the state's 2024 AI Act and replaces it with a narrower, notice-based law that takes effect January 1, 2027. If you use automated decision-making technology in a "consequential decision" about a Colorado resident — employment, education, housing, financial services, insurance, health care, or essential government services — you have to give clear notice before you use it. If the decision goes against someone, you owe them a plain-language explanation within 30 days, a way to see and correct their data, and meaningful human review. The Colorado Attorney General enforces it, there's a 60-day window to fix most problems, and there's no private right to sue.
Most small businesses aren't making those decisions with AI. If you screen job applicants with an AI tool and you hire in Colorado, you might be.
European Union: the AI Act, delayed on its high-risk rules. The EU's amendment — the Digital Omnibus on AI — was published on July 24, 2026 and took effect July 27. It pushed the high-risk rules back: stand-alone high-risk systems like hiring or credit scoring now start December 2, 2027, and AI built into regulated products starts August 2, 2028. What didn't move: the transparency rules started August 2, 2026 as planned. If people in the EU talk to your chatbot, it has to be clear they're talking to AI. Systems already on the market before August 2 have until December 2, 2026 for the content-labeling piece. None of this touches you unless you sell into Europe.
| Item | What it is | Status, September 27, 2026 | |---|---|---| | New York City AI package | Ten bills, including a human-override kill switch and outside validation for AI sold or used in the city, 24-hour incident reporting for city contractors, and paying whistleblowers a share of fines | Introduced September 25. Hearing October 5. Proposed. | | Ban Artificial Superintelligence Act | Senator Sanders and Representative Casar's federal bill to ban superintelligent AI, pause advanced development until a new Department of AI writes rules, with prison terms up to 20 years | Introduced September 23. Proposed. | | California executive order | Newsom's September 18 order tells a state agency to speed up SB 813 and AB 1405 and bring back recommendations within two months, including a shutdown mechanism for frontier models | An order to study and recommend. Not a statute. |
A free account opens the rest of this article and the whole library.