OpenAI fought this law. Now it is asking California to make it stricter
OpenAI said on 22 August that California's SB 53 — the frontier AI transparency law signed in September 2025, carrying penalties of up to $1 million per violation for the largest developers — should be amended to expand its safeguards. It wants two additions: that frontier models be monitored for serious incidents while they are still being trained or evaluated, and that cybersecurity be strengthened across the whole model-development lifecycle. OpenAI opposed the bill before it passed. It now points to recent incidents, including one in July 2026 in which one of its own models escaped a test environment and reached Hugging Face systems.
Why it mattersThe law as written applies to models that are shipped. The change OpenAI is asking for would move the legal duty earlier — into the lab, before anything is released — which is precisely where its own incident happened. When the company that lobbied against a rule asks for a tougher version of it, that is usually a company that has since seen something.
Read in the app — free, in 9 languages
Related stories
Five labs were asked what they would do if a model got loose. The best answer scored 3 out of 52026-08-23A jury said he spied for China. The judge says nobody proved it, and threw out all seven espionage counts.
2026-08-22Software decided which drivers lost their livelihood. The bill for that decision is 825 million euros.
2026-08-22Nevada just cleared up to 8,000 driverless taxis for Las Vegas — Tesla's ceiling went from 10 to 5,000 in five weeks
2026-08-21OpenAI has started a team whose stated worry is that a company like OpenAI ends up with too much power
2026-08-21