By Ed Eisenstein
New York is going to move forward with one of the most ambitious attempts to regulate advanced artificial intelligence (AI). Under the Responsible AI Safety and Education (RAISE) Act, major developers of powerful frontier AI models will face new registration, transparency, safety and incident-reporting requirements.
But there is a fundamental question behind all of this: How exactly does a state regulate something as complex, distributed and rapidly evolving as AI?
The answer is that New York isn’t really regulating AI itself. It is regulating the companies that build it. That distinction matters.
You Can’t Put a Regulator Inside a Neural Network
Modern AI systems can contain hundreds of billions of parameters and exhibit capabilities that aren’t always obvious to their developers before extensive testing.
New York can’t put an inspector inside a neural network to watch bad capabilities emerge.
The RAISE Act uses a more traditional regulatory approach: require companies developing extremely powerful AI systems to establish safety procedures, document what they are doing, disclose certain information and report serious incidents.
Covered developers must establish and publish safety and transparency frameworks explaining how they evaluate potentially catastrophic risks and what safeguards they use. If a qualifying critical safety incident occurs, the developer must report it to the state within 72 hours. The state can then enforce those requirements through the legal system, including significant civil penalties for certain violations.
In other words, New York isn’t controlling AI. It is creating legal accountability for the humans and corporations developing it.
But How Does New York Know What AI Is Actually Doing?
This is where regulation becomes considerably more difficult. Suppose researchers discover during testing that an advanced AI model has developed a dangerous capability. Unless that capability becomes publicly visible, regulators may have no independent way of knowing that it exists.
The regulatory system depends substantially on developers testing their own systems, maintaining records, following their safety frameworks and reporting qualifying incidents.
That creates an obvious oversight problem: Who verifies the people doing the verifying?
New York lawmakers are already considering that question. Legislation introduced in 2026 proposes requiring annual independent third-party verification of whether major AI developers are complying with their safety frameworks.
Independent auditing could strengthen oversight, but it still doesn’t solve the underlying technical problem of completely understanding what happens inside highly complex AI systems.
AI Doesn’t Stop at the New York Border
There is another major limitation: artificial intelligence is global.
An AI model can be developed in California, trained using data centers in another state, operated from Europe and accessed by someone in New York within seconds.
Open-source models create an even more complicated problem. Once model weights are publicly distributed, thousands of organizations and individuals around the world can potentially modify and operate them.
A state government has considerable authority over companies operating within its jurisdiction. It has far less ability to control software distributed around the world.
What New York Can Actually Control
New York does have several powerful tools.
It can require companies subject to its laws to maintain records and make disclosures. It can investigate suspected violations. It can impose financial penalties. It can establish reporting requirements. It can require companies to implement particular governance processes.
Those are meaningful regulatory powers. But they shouldn’t be confused with technical control over AI itself.
There is no master switch in Albany that allows the government to shut down artificial intelligence. There currently is no technological mechanism that gives regulators full visibility into what every sophisticated AI model has learned or what capabilities it might develop.
Whether mandatory safety frameworks, incident reporting, independent testing, audits and financial penalties ultimately make advanced AI substantially safer remains an open question.
And that may be the most important point in the entire debate.
Artificial intelligence is advancing faster than traditional regulatory systems normally operate. Governments are attempting to construct oversight mechanisms while the technology itself continues to change.
New York’s experiment will, therefore, be important to watch — not because the state has discovered how to control artificial intelligence, but because it is attempting to answer a much more realistic question: When nobody can completely control AI, how do we make the people building the most powerful systems accountable for what happens?
Ed Eisenstein is the CEO of United Network Associates, a cloud service and cybersecurity provider based in Farmingdale.