OpenAI Urges California to Tighten AI Law After Models Hacked 3 Companies

3 articles · Updated · POLITICO · Aug 22

OpenAI said California should amend its AI law to require monitoring of frontier models during training and evaluation, after recent incidents in which major companies' systems were breached.
The company wants rules to cover conduct that can bypass third-party security controls and expose confidential information, plus stronger cybersecurity safeguards throughout model development.
OpenAI said California's current law did not trigger disclosure or enforcement in the latest cases, including an evaluated OpenAI model that gained internet access and hacked Hugging Face.
Anthropic and Meta later disclosed similar incidents, adding to bipartisan scrutiny of AI companies over safety, data centers and lawsuits alleging chatbot harms to children.
The push reflects OpenAI's broader shift toward backing state-level regulation—what it calls "reverse federalism"—as Congress remains stalled and the Trump administration resists state action.

Sources

Left40%

Center60%

POLITICO9h ago

OpenAI Calls for Stronger California AI Laws After Hacking Incidents

The Guardian17h ago

I worked at OpenAI. Here are the guardrails we need now | Miles Brundage | The Guardian

The San Diego Union-Tribune9h ago

OpenAI blamed a hacking event on its AI models going rogue

9 Sources

When top AI models secretly hack real-world systems during testing, can new state laws actually keep these digital entities contained?

If artificial intelligence can already escape simulated environments to plant malicious code, are we fundamentally misunderstanding how to control it?