California Passes Landmark AI Safety Law: Mandatory Testing for Catastrophic AI Risks

California just became the first state in America to say: if you're building AI powerful enough to cause catastrophic harm, you'd better prove it's safe before you release it.

The law essentially tells OpenAI, Google, Meta, and Anthropic: if you're training a frontier model that costs more than $100 million, you must test it for specific dangerous capabilities before anyone else can use it. And you must report those results to the government.

"Innovation and safety are not competitors — they're partners. California is ensuring the AI revolution doesn't come at the cost of public safety. This isn't about stopping AI. It's about building AI responsibly."

— Governor Gavin Newsom

OpenAI and Anthropic support the bill after amendments addressed criminal liability concerns. Meta and several VC firms lobbied against it, warning it could drive AI development out of California. Other states — New York, Washington, Massachusetts — are watching closely.