California Just Made the First Real Move on AI Safety — What Comes Next
Gov. Gavin Newsom's executive order on AI safety is the most aggressive state-level move yet, but its real test is whether the rest of the country follows California's lead — as it always does.
California’s ‘Urgent Velocity’ Play
Governor Gavin Newsom didn’t mince words when he signed the executive order Friday. The federal government, he said, has shown “abject failure” on AI oversight, and California is stepping into the breach with what he called “thoughtful but urgent velocity.”
The order itself is blunt: a group of experts gets two months to produce a guide for strengthening the state’s AI safety and security laws. The announced possibilities include requiring independent third parties to write safety plans for frontier AI companies and mandating that those companies develop a “kill switch” — a mechanism to shut down their models in an emergency.
This is not symbolism. It’s a template. And if history is any guide, the rest of the country will follow California’s lead whether Washington likes it or not.
The Anthropic Spark
The policy move landed hard on the heels of an Anthropic researcher resigning this month after posting publicly about AI’s potentially humanity-ending capabilities. The post set off alarm bells in Washington, where lawmakers have been circling the issue for months but producing nothing but committee hearings and press releases.
Congress appears unlikely to take meaningful legislative action before the 2026 midterms. President Donald Trump has dismissed AI safety concerns as a “hoax.” That vacuum is exactly what Newsom is exploiting — and filling.
The politics are layered. Newsom hasn’t formally launched a 2028 presidential campaign, though he’s acknowledged he’s considering it. He’s joined a cohort of Democratic governors and senators — Pennsylvania’s Josh Shapiro, Arizona’s Mark Kelly, New Jersey’s Cory Booker — who are making AI regulation a central plank. Shapiro has called for third-party oversight of frontier models. Kelly is pushing legislation for an AI-displaced-worker fund. Booker wants Trump to convene a special congressional session on AI risks.
What connects them all is a simple calculation: in a country where federal action is gridlocked, state-level leadership is the only way to move the issue forward — and the political capital that comes with it.
The Kill Switch Question
A kill switch mandate is the kind of policy that sounds decisive until you ask engineers what it actually means. Shutting down a large language model mid-inference is technically possible in narrow contexts, but requiring every frontier AI company to build one — and defining what triggers deployment — is far messier. Who decides an emergency exists? Is it a human operator, an automated system, a regulator?
That ambiguity is both the order’s weakness and its strength. The vague language lets it pass. The detail will matter later, when companies like Anthropic, OpenAI, and Google face compliance deadlines and legal challenges.
What’s clear is that California is already building the institutional scaffolding. Earlier this month, Newsom signed legislation creating a framework for third-party AI auditors. The executive order builds directly on that foundation. This isn’t a one-off press event — it’s a rolling regulatory architecture.
California’s Shadow Standard
The deeper story here isn’t about AI safety in California. It’s about whether state-level tech regulation becomes the de facto national standard again.
California has done this before. The California Consumer Privacy Act set the privacy benchmark for the entire country. The state’s cab driver classification rules forced Uber and Lyft to change their employment structures nationwide. The Chinese-language media missed this pattern because it’s not headline-grabbing — it’s structural.
If California’s AI framework becomes the default compliance standard for any company doing business in the state — and that’s every major AI developer — then Washington’s inaction simply becomes a delay, not a blockade. Companies will build to California’s rules. Other states will adopt similar frameworks. The market does what Congress won’t.
That’s the implicit argument in Newsom’s closing line: “California has already built a national model, and our policy should be the national baseline.”
He’s not being naive. He’s describing a process that’s already repeated itself a dozen times.
Who Wins, Who Loses
The winners are regulators who wanted action and couldn’t get it federally. They now have a laboratory. The policy will be stress-tested in the real world, refined through litigation and implementation, and exported to other states.
The losers are AI companies that preferred self-regulation or no regulation at all. They’ll have to budget for third-party audits, safety plan compliance, and kill switch infrastructure — costs that compound across jurisdictions if other states adopt different versions. The regulatory fragmentation risk is real: California’s kill switch definition could conflict with Texas’s or New York’s.
The biggest loser is probably the American public’s expectations of federal governance. When a governor can outpace Congress on a technology that’s reshaping the economy in real time, the question stops being whether California is overreaching and starts being why Washington isn’t.
What Happens Next
The two-month deadline for the expert guide is the first real checkpoint. Whatever emerges there — precise language, enforceable requirements, exemptions — will set the trajectory. If the guidelines are strong, industry will push back in court. If they’re soft, the movement for federal action accelerates.
The 2026 midterm cycle will be the political inflection point. Any Democratic candidate who can point to a working state-level AI safety regime has a concrete governing credential. Any Republican who opposes it faces a choice: side with tech donors or side with voters who want guardrails.
And Trump’s continued dismissal of AI safety — calling it a “hoax” — places him further from the center on this issue than on almost any other technology policy question. That gap is widening.
Newsom’s order doesn’t solve AI risk. It solves the political problem of doing nothing while the technology advances faster than any regulatory framework can track. In that sense, it’s not the endgame — it’s the opening move.
The real test is whether California’s framework survives contact with the companies it’s meant to regulate, and whether Washington finally decides to meet it halfway.