What does the new California law do?

It creates a framework for Independent Verification Organizations that can test frontier AI models before they are released.

Why is AI oversight considered difficult?

The models are so vast that they require other AI models to analyze them, and investigations can be expensive, like the METR probe into an OpenAI-Hugging Face hack that cost $400,000 in AI tokens.

AI Regulation & PolicyAI Safety & AlignmentSemafor Tech53m agoSep 5, 2026, 06:03 JST1 min read

California law mandates AI testing before release

Send on LINE

Key takeaway

California passed a new law. It creates testing bodies for frontier AI models.
The law aims to reduce risks before release.
Yet AI oversight may be impractical and costly.

3 Key Points

What happened
California passed a law creating Independent Verification Organizations to test frontier AI models before release, aiming to limit risks.
Why it matters
AI oversight is questioned as impossible and too costly; models are so vast that other AIs are needed to analyze them, and one hack investigation cost $400,000 in AI tokens.
What to watch
Whether nimble governance can replace laws that become outdated quickly, as the article suggests.

Ask the AI about this article →

Summaries like this, in your inbox every morning.

Get news like this every morning — free

Context & Analysis

The law arrives as doubts grow over whether AI oversight is feasible. The METR investigation shows practical challenges: significant costs and technical hurdles. The article argues for adaptive governance rather than fixed rules, suggesting that smart people should create rules open to correction. This approach aims to prevent ineffectual or counterproductive laws.

FAQ

What does the new California law do?: It creates a framework for Independent Verification Organizations that can test frontier AI models before they are released.
Why is AI oversight considered difficult?: The models are so vast that they require other AI models to analyze them, and investigations can be expensive, like the METR probe into an OpenAI-Hugging Face hack that cost $400,000 in AI tokens.

Semafor TechRead Original Article

Send on LINE