New York just became the first state to require frontier AI developers to open their books. And the clock is ticking.

Governor Kathy Hochul signed Senate Bill 8828 this week, overhauling the Responsible AI Safety and Education Act (the RAISE Act) she originally signed in December 2025. The original version was vague enough that Hochul herself criticized it, writing in a memo that it would "impose broad compliance obligations on large-scale models without adequate specificity." The rewrite fixes that by getting very specific about what frontier AI companies must do.

The law takes effect January 1, 2027. Here is what it demands.

The rules

Frontier model developers operating in New York must publish detailed transparency reports covering model capabilities, known limitations, safety testing results, and training data composition. The law creates a new office within the New York Department of Financial Services (DFS) with rulemaking authority to define specific reporting formats and additional requirements.

Governance requirements include designated safety officers, incident response processes, and mandatory risk assessments before deploying new model versions. The DFS office can conduct investigations and impose penalties for noncompliance.

Translation: if you are OpenAI, Anthropic, Google, or Meta, and your model is available to New York residents (which means everyone), you now have a compliance deadline.

Why this is different

Other states have passed AI laws. Colorado regulated algorithmic discrimination. California tried and failed to pass SB 1047. Texas and Illinois have targeted deepfakes and biometric data.

New York's law is different because it goes after the model layer directly. It does not just regulate what you do with AI. It regulates what the AI itself is. Training data transparency. Capability disclosure. Safety testing results. This is the kind of regulation that frontier labs have been lobbying against at the federal level for two years.

And it arrived in the state where most of these companies have their largest office outside San Francisco.

The federal preemption problem

The Trump administration has been pushing hard to preempt state AI laws with a federal framework. The White House published its own AI policy blueprint last month, and multiple bills in Congress would override state-level regulations.

New York is betting that federal action will not arrive before January. Given that Congress has introduced 47 AI bills and passed zero, that bet looks solid.

The result is a patchwork that AI companies have been dreading. New York wants transparency reports. Colorado wants bias audits. The EU wants risk classifications. China wants content controls. Every frontier lab now needs a compliance team the size of a small law firm just to keep track of which jurisdiction requires what.

What to watch

The DFS rulemaking process starts immediately. The specifics of what counts as a "frontier model" under the law, what the reporting format looks like, and what penalties apply will be defined through administrative rulemaking over the next eight months.

Every major AI company should be sending lawyers to Albany. If they are not already, they will be by the time the DFS publishes its first proposed rules.