Illinois just became the latest state to pass sweeping AI safety legislation, and it’s notable for two reasons: the Trump administration can’t do much about it, and the companies being regulated aren’t fighting back.

The Illinois Artificial Intelligence Safety and Transparency Act, signed into law this week, requires AI companies deploying models in the state to conduct safety testing before release. That includes evaluations for risks like bias, security vulnerabilities, and potential for misuse. Companies must also register with the state and submit testing documentation annually.

What makes this different from most tech regulation is that Anthropic and OpenAI both publicly supported it. That’s unusual. These companies have spent years pushing back on safety testing requirements at the federal level, arguing they’re too burdensome or technically infeasible.

So what changed? The Illinois law is narrower and more practical than earlier proposals. It doesn’t ban specific AI capabilities or impose criminal liability on developers. Instead, it creates a testing framework that major labs say they’re already doing internally. Anthropic’s chief policy officer called it “a model for other states” in a statement. OpenAI’s head of policy said the law “aligns with how we already develop models.”

That’s partly true. Anthropic and OpenAI both run internal red-teaming exercises and publish some safety evaluations. But the Illinois law makes those tests mandatory and creates consequences for skipping them. Companies that don’t comply face fines up to $50,000 per violation, and the state attorney general can bring enforcement actions.

Why Trump Can’t Stop This

The timing matters. The Trump administration has spent the past year trying to roll back Biden-era AI policy, including executive orders that encouraged voluntary safety commitments. Trump’s AI czar has called safety testing “innovation theater” and pushed agencies to focus on promoting U.S. competitiveness instead of regulating risk.

But state laws operate independently. Illinois doesn’t need federal permission to regulate AI companies doing business in the state, and Trump can’t preempt state safety requirements without Congress passing a law that explicitly blocks them. That’s not happening anytime soon. Republicans in Congress don’t agree on AI policy, and Democrats have no interest in helping Trump weaken state regulations.

This creates a patchwork problem for AI companies. Illinois is the third state to pass comprehensive AI safety legislation this year, following California and New York. Each law has slightly different testing requirements and compliance deadlines. Colorado is considering similar legislation. So is Massachusetts.

For companies like Anthropic and OpenAI, supporting a reasonable state law now might be strategic. If they help shape the first wave of state regulations, they can push for consistency across states and avoid dealing with 50 different testing regimes. It’s also better public relations than fighting every bill and losing anyway.

What the Law Actually Requires

The Illinois law applies to “high-risk AI systems,” which it defines as models that could meaningfully affect decisions about employment, housing, credit, education, healthcare, or criminal justice. It also covers models that interact directly with the public and could cause physical or economic harm.

That’s a wide net, but the law includes carve-outs. It doesn’t apply to AI used for basic tasks like spell-checking or spam filtering. It also exempts models that only process data internally and don’t interact with Illinois residents.

For covered systems, companies must conduct testing in four areas: algorithmic bias, cybersecurity risks, ability to generate harmful content, and potential for misuse. They need to document the testing process and results, then submit a compliance report to the Illinois Department of Innovation and Technology.

The law also requires companies to disclose when people are interacting with AI, not humans. That’s aimed at chatbots and customer service systems. There’s an exception for obvious cases, like when you’re clearly talking to a search engine.

Enforcement starts in January 2027. The state will publish detailed testing standards by September, which gives companies about eight months to get compliant.

The Real Test Is What Happens Next

Illinois isn’t the first state to regulate AI, but it might be the first to do it in a way that sticks. California’s earlier AI safety bill died in committee after opposition from tech companies. New York’s law focuses narrowly on hiring algorithms, not general-purpose models.

The Illinois law is broader and has industry support, which makes it harder to ignore. If the testing requirements turn out to be workable and companies actually comply, other states will copy the framework. If it turns into a bureaucratic mess or companies find easy workarounds, the political will for state AI regulation will weaken.

What’s clear is that the locus of AI policy is shifting from Washington to state capitals. Trump’s deregulatory push means there won’t be meaningful federal AI safety legislation in the near term. That leaves states free to experiment, and companies have to decide whether they want to fight 50 separate battles or accept some baseline testing requirements.

For now, Anthropic and OpenAI are choosing acceptance. Whether that’s because they genuinely believe in safety testing or because they’ve decided they can live with it doesn’t really matter. The law exists either way, and it will affect how AI gets deployed in the fifth-largest state by population.

Trump can complain about it, but he can’t stop it. That might be the most important part of this story.