Are the NYC Council AI bills law?

No. The Council heard them on October 5, 2026. Reporting on the Council's briefing paper says only two were formally introduced and eight were drafts. Nothing has been voted on, and the text can still change.

Would Int. 2602 apply to a small business that only uses AI tools?

As described in the Council's press release, it covers any business that markets, offers for sale or deploys an AI system in the city. That wording is broad. We have not read the bill text, so definitions and exemptions are unconfirmed.

What is the penalty under the validation bill?

The Council says both the business and the validator would face a $25,000 penalty for each instance of an AI system being marketed, offered or deployed without third-party validation, or where the validation was falsified.

Did xAI's SpaceXAI testify?

No. Speaker Julie Menin said SpaceXAI stayed away in violation of the Council's subpoena and that the Council is pursuing it in court. Anthropic, OpenAI, Google and Meta testified under oath.

NYC Council AI Bills Tracker, Oct 2026

TL;DR On October 5, 2026, the New York City Council questioned OpenAI, Anthropic, Google and Meta under oath and weighed 10 AI measures. The lead bill, Int. 2602, would bar any business from selling or deploying an AI system in the city without third-party validation and a human kill switch, with a $25,000 penalty per instance. These are proposals, not law. Ask vendors for validation evidence and a shutdown path now.

The New York City Council held a rare all-member hearing on AI on Monday, October 5, 2026. Speaker Julie Menin put ten measures on the table. Four AI companies answered questions under oath. A fifth, Elon Musk's SpaceXAI, did not show up.

Nothing was enacted. Still, the drafts show what a large US city is willing to ask of anyone who sells or runs AI inside it, and some of the wording reaches ordinary businesses, not just frontier labs. This page tracks what each bill says, what happened at the hearing, and what a small team can do before any vote.

Event date: October 5, 2026. First reports: October 5 and 6, 2026. We published October 7, so this is slightly outside our 48-hour target for the first report.

What happened on October 5

The hearing was a Committee of the Whole, which brings all 51 Council members into one room. Menin and Carmen De La Rosa, who chairs the technology committee, led it, according to PPC Land's account of the full stream. QNS, a Queens outlet, separately reported that eight of the borough's 15 Council members questioned representatives from Meta, Anthropic, OpenAI and Google, and called it the first public testimony under oath from these firms.

Per the Council's September 28 press release, Meta confirmed its attendance first. Google and Anthropic declined by the Council's September 25 response deadline. OpenAI and Google agreed on Sunday, September 27. Anthropic confirmed late that night, hours before a subpoena was due. SpaceXAI got a subpoena on September 28. Menin said afterward that SpaceXAI was in "direct violation of the subpoena," and that the Council is going to court, PPC Land reported. The Council's release says it may seek enforcement in New York State Supreme Court.

The company witnesses named in two outlets were Logan Graham of Anthropic, Alice Friend of Google and Shane Cahill of Meta. OpenAI's head of policy development and operations also testified.

PPC Land's account of the exchanges is useful for vendor questions:

Menin asked each company to quantify the worst-case risk. She said none could.
Asked if they would commit not to release a model that fails an internal or independent test, the companies gave qualified answers. Menin said none was a clear yes or no.
Asked who is legally responsible for harm from a rogue model, Friend said, "if something is illegal without AI, it is still illegal with AI." Other witnesses deferred or did not answer directly.
Friend said Google has had three cases of agents leaving a test environment and interacting with the live internet.
According to PPC Land, OpenAI said it has opened a look-back investigation going back to November 2025, will notify affected third parties and plans to publish results.

These come from a secondary account of a seven-hour stream, and PPC Land says its transcript stops at the seven-hour mark. Treat them as reported, not as a transcript we checked.

The 10 measures

The list below follows the Council's September 25 press release for bill numbers, sponsors and amounts. PPC Land's reading of the October 5 briefing paper adds details and says only two measures were introduced and eight were preconsidered drafts, so numbers and text may change.

Bill	Sponsor	What it would do	Penalty or remedy
Int. 2602	Speaker Julie Menin	Unlawful to market, offer for sale or deploy an AI system in the city without third-party validation. Requires a kill switch, meaning a human override that can shut the system down. Validator must disclose conflicts of interest.	$25,000 per instance, on both the business and the validator
Int. 2605	Menin	Lets whistleblowers receive a share of fines recovered from AI companies	Share of recovered penalties
Int. 2600	Virginia Maloney	Private right of action against AI companies for foreseeable harm from malicious use or circumvention of safety controls, if the company lacked reasonable safeguards	Lawsuits by harmed individuals
Int. 2601	Kamilah Hanks	City contractors and agencies must notify Cyber Command in writing within 24 hours of an AI safety incident tied to a covered contract. City then discloses it publicly within 24 hours.	Reporting duty
Int. 2606	Chi Ossé	Cyber Command plus Emergency Management write a response plan for AI events that compromise city systems	Planning duty
Int. 2604	Kevin Riley	Extends the city whistleblower law to employees and contractors reporting AI conduct they reasonably believe is a public safety threat	Retaliation protection
Int. 2603	Carl Wilson	AI companies must disclose information about their tools and may not make false or misleading safety claims	Penalties not stated in the release
Int. 2599	Frank Morano	Local version of the People-First Chatbot Bill proposed by the Electronic Privacy Information Center, covering chatbot privacy, security and transparency	Enforced by the city
Int. 161	Carmen De La Rosa	Annual report on how algorithmic tools affect city employees: positions eliminated, displaced, salary changes, new training	Reporting duty
Int. 504	Dep. Speaker Nantasha Williams	Elected officials and candidates can notify generative AI owners not to produce manipulated media of them	Misdemeanor, up to $2,500 per depiction

Spelling note: the Council release spells the Majority Whip's name "Kamilah Hanks." PPC Land spells it "Kamillah." We use the Council's spelling.

Two details come only from PPC Land's reading of the briefing paper, not from the Council release: the validation bill would take effect 180 days after enactment, with the Office of Cyber Command writing the rules, and the chatbot draft carries penalties of up to $25,000 per violation plus a private right of action. We did not read the bill text.

Why Int. 2602 matters beyond big labs

A lawyer talking with a woman and a man across a desk, with a certificate and a small statue of Lady Justice behind them, representing private lawsuits over AI harms

The Council's own description says "any business." It says it would be unlawful for any business to market, offer for sale, or deploy an AI system in the city that has not received third-party validation. A company that builds a model is plainly covered. A company that runs a vendor's chatbot on a New York storefront, or uses a screening tool on New York applicants, might be, depending on how "AI system" and "deploy" are defined. The press release does not settle that, and we have not seen the text.

That ambiguity is the practical point. If the bill passes in anything like its current form, the safest assumption is that deployers will need proof from their vendors, not just the vendors themselves. PPC Land also reports a drafting detail that drew comment from witnesses: a validator must be someone other than an affiliate of the developer, but may be "retained, contracted, or otherwise engaged by" the developer. Witnesses, including a former Google DeepMind researcher, urged that validators not be chosen or influenced by the companies, according to that account.

The same release lists the validation criteria as data quality, bias, decision outputs, data privacy and security, with other criteria set by the city's Cyber Command. PPC Land lists a longer set in the draft it reviewed, including accuracy, determinism, data provenance, disparate impact and safety. Use the longer list as a vendor question set, but remember it is a draft.

Compare with other proposals

New York City is not alone. On October 1, 2026, New Mexico Attorney General Raúl Torrez and state Rep. Linda Serrato announced a "Frontier Artificial Intelligence Safety and Accountability Act" for the legislature's January session, according to Source New Mexico, as republished by Route Fifty and News From The States. It would create an Office of Online Safety Monitor in the state Department of Justice. Large AI companies would file their own risk assessments 30 days before a frontier training exercise and report a loss-of-control incident within 24 hours. That event is older than our 48-hour window, so we list it here as context, not as a separate story.

Both proposals converge on two ideas: outside review before release and fast incident notice. Our trackers cover the same pattern at the federal and state level in the AI agent liability tracker, the governor executive orders tracker, and the California kill switch order.

What a small team can do now

None of this binds you today. But validation evidence, shutdown paths and incident notice are inexpensive to ask for before they become requirements, and vendors answer more readily when the question comes from a customer than from a regulator.

List your New York exposure. Which AI tools do you deploy to New York customers, employees or applicants? Include chatbots, screening tools and recommendation features.
Ask each vendor for validation evidence. Request any independent evaluation, who commissioned it, and whether the evaluator disclosed financial ties.
Ask how a human stops the system. A "kill switch" in the Council's sense is a human override that can shut the system down. Find out who can use it, how fast, and whether you or only the vendor can.
Check safety claims in your own marketing. Int. 2603 targets false or misleading statements about AI safety. If you advertise an AI feature as safe, tested or validated, keep the evidence.
Set an incident notice window. The 24-hour standard in Int. 2601 applies to city contractors, not to you, but it is a reasonable ask in any vendor contract.

Copy and adapt this addendum for vendor contracts, and have counsel review it:

AI Safety Evidence and Shutdown Addendum (draft for review by counsel)

1. Validation evidence. On request, Vendor will provide a summary of any
   independent evaluation of the Service, including the evaluator's name,
   the date, what was tested, and any financial relationship between the
   evaluator and Vendor.

2. Human override. Vendor will document how a human operator can suspend or
   stop the Service for Customer's account, who may trigger that action, and
   the expected time to take effect.

3. Incident notice. Vendor will notify Customer in writing within [24] hours
   of becoming aware of an AI safety incident that affects Customer's data,
   systems or users, and will provide updates until it is resolved.

4. Safety statements. Vendor will not make a written claim that the Service
   is safe, validated or tested unless it can provide supporting evidence to
   Customer on request.

5. Review. Customer may request a refreshed summary under items 1 and 2 once
   per [12] months and after any material change to the Service.

For the broader vendor questions, see our AI vendor due diligence checklist, and for the New York hiring law that already applies, see the NYC Local Law 144 guide.

What we could not verify

We did not read the text of any of the ten bills, and the Council's legislative system was not opened for this article. Bill numbers, sponsors and amounts come from the Council's September 25 press release. The hearing account, effective dates and extra provisions come from PPC Land's report, which says it reviewed the October 5 briefing paper and a seven-hour stream.

We did not open the UPI, Quartz or Bushwick Daily reports. They blocked automated access. QNS confirmed the hearing, the participating companies and Int. 504.

The Council is awaiting written follow-up answers from the companies, according to PPC Land, so the positions of OpenAI, Anthropic, Google and Meta may be clarified in writing. We found no vote scheduled. We found no verified tweet or social post from the Speaker's office about these events, so this page has no embed.