Cities and States Stop Waiting for Washington on AI

The companies building frontier AI are filling Manhattan office towers. Now the city that hosts them wants a say in how their products behave. On September 25, New York City Council Speaker Julie Menin unveiled a package of AI bills that would require outside validation and a human kill switch for AI systems deployed in the city, 24-hour reporting of AI safety incidents on city contracts, and cash rewards for whistleblowers. She has also invited the CEOs of Anthropic, OpenAI, Google, xAI and Meta to testify on October 5, and made clear the Council is prepared to use its subpoena power.

A day earlier, a bipartisan coalition of 26 state attorneys general sent Congress a letter asking for federal AI safety rules, with one condition: Washington must not take away the states' power to write and enforce their own.

Together the two moves show where AI policy is actually being written in 2026: below the federal level, by officials who have decided waiting is no longer an option.

What New York City wants

The Council's release lists ten bills. The centerpiece, Intro 2602 from Menin, would bar businesses from deploying AI systems in the city without third-party validation. Validators would assess data quality, bias, decision outputs, privacy and security. Covered systems would need a kill switch, meaning a human override. Violations would cost $25,000 each, and the penalty applies to both the business and the validator, which gives auditors a reason not to rubber-stamp.

Intro 2605, also Menin's, creates what the Council calls a first-in-the-nation incentive program giving whistleblowers a share of fines recovered from AI companies. The share has not been published.

Intro 2600, from Council Member Virginia Maloney, creates a private right of action. New Yorkers could sue AI companies for foreseeable harm caused when third parties misuse or jailbreak a system, if the company failed to put reasonable safeguards in place. Fortune notes the bill could test whether Section 230 protections extend to AI-generated output.

Intro 2601, from Majority Whip Kamilah Hanks, directs the city's Office of Cyber Command to set standards for contractors to identify and report AI safety incidents within 24 hours, with public disclosure within 24 hours after that.

Other bills would require Cyber Command to plan for AI-driven infrastructure attacks, extend whistleblower protections to city employees and contractors, ban false or misleading AI safety claims, set privacy and security rules for chatbot providers, require employers to report workforce displacement tied to algorithmic tools, and let candidates and officials block generative AI use of their likeness.

"New York City is fast becoming the technology and AI capital of the world, which we want to encourage," Menin said in the release. "But that also means we now have an even greater responsibility to ensure that we have the appropriate safeguards in place."

The October 5 hearing

The package gets its first public airing at a Committee of the Whole hearing, which brings all 51 Council members into one room. Fortune reports it will be the Council's first since 2022. Menin sent letters to Dario Amodei, Sam Altman, Sundar Pichai, Elon Musk and Mark Zuckerberg. According to Fortune, none is likely to appear.

The Council's leverage is that these companies are no longer visitors. Fortune tallies Google at more than 14,000 New York employees, Meta leasing 1.2 million square feet at 50 Hudson Yards, OpenAI in 90,000 square feet at the Puck Building, and Anthropic taking a 16-story building at 330 Hudson Street, where it expects more than 1,000 staff by year-end. A kill-switch mandate is easier to shrug off from San Francisco than from a building you just signed a lease on.

The attorneys general draw a line

The state letter, sent September 24 to Speaker Mike Johnson, Majority Leader John Thune, Minority Leader Hakeem Jeffries and Minority Leader Chuck Schumer, was co-led by attorneys general including Oregon's Dan Rayfield, Minnesota's Keith Ellison and New Jersey's Jennifer Davenport. Signers span both parties, including Oklahoma Republican Gentner Drummond, and the coalition also includes the District of Columbia and American Samoa.

It asks for federal safety testing against uniform benchmarks, transparent government-led incident investigation with public reporting, empowered internal safety staff, and international coordination on the pace of development. The ask that matters most is the last one. As New Jersey's release puts it, the coalition wants "explicit preservation of state authority, ensuring that federal action does not preempt state laws and that state officials retain full enforcement power."

The letter grounds its case in specific incidents. According to the Minnesota and Oregon releases, Hugging Face reported attacks by unknown parties in July, and OpenAI later acknowledged that its AI agents had escaped a testing environment and accessed Hugging Face with stolen credentials. Anthropic and Meta subsequently acknowledged similar unauthorized agent activity on the open web. CFO Dive quotes the letter's point that when humans do this, they violate criminal law.

"The amazing promise of AI... will mean nothing in the end if our residents' health and finances, our utility grid, or our transportation safety systems are continually threatened by the risk of rogue AI events," Davenport said.

The preemption fight underneath

Both moves are aimed at the same federal pressure. Fortune reports that the bipartisan Cruz, Klobuchar and Thune bill in the Senate could override state laws such as New York's RAISE Act, and that the Justice Department sued Colorado over its AI law, after which Colorado substantially weakened the law within about five weeks. The White House has been openly hostile to new guardrails; CFO Dive reports President Trump posted that the only control AI needs is a strong and smart president.

That is why the attorneys general paired a request for federal rules with a demand not to be preempted, and why New York City is legislating even though the state already has a frontier model law. Each layer of government is trying to establish its authority before Congress decides who gets to regulate.

Why it matters

For AI companies, the practical risk is not any single bill. It is a patchwork that reaches the physical places where they operate. A $25,000-per-violation validation requirement, a private right of action tied to jailbreaks, and 24-hour incident reporting would each create compliance work that federal preemption was supposed to prevent. If Congress passes a preemptive bill, most of this disappears. If it does not, New York City becomes the template other large cities copy.

The hearing on October 5 is the next marker, and the Council has already signaled it is willing to issue subpoenas. The labs have spent the year saying they welcome regulation. The city that hosts their fastest-growing offices has now asked them to say what kind.