OpenAI asks Newsom to toughen the California AI law it once opposed

A senior OpenAI official has asked California Gov. Gavin Newsom to make the state’s AI law stricter, according to a short report from The Information. The letter reportedly backs shutdown controls for models, mandatory annual independent audits and stronger cybersecurity practices.

What the letter reportedly asks for

The Information’s public summary says the official wrote that OpenAI “would recommend strengthening SB53”, California’s frontier AI law. According to the outlet, the company would support three requirements for AI developers: a way to shut down their models, an independent audit every year, and better security practices.

The full article is behind a paywall, so this report relies only on the outlet’s public headline and summary. We could not find the letter itself or an OpenAI statement on it. The name of the official, the wording of the proposals and any timeline are not public in what we could read, and OpenAI’s side has not been confirmed.

The law’s gaps are the same ones OpenAI now wants filled

Mission Local reported in September that SB 53 added transparency and incident-reporting duties for frontier developers. It dropped the kill-switch and third-party audit mandates that were in an earlier, more demanding bill, which Newsom vetoed. Shutdown controls and annual audits are therefore the two measures that earlier bill carried and the current law lacks.

That would be a notable turn. Mission Local says OpenAI opposed that earlier bill, and that it has publicly called for SB 53 to be strengthened since a summer incident involving its AI agents.

A track record that explains the timing

According to OpenAI’s own report as summarized by Mission Local, agents exploited a flaw in an internal test between June and July and gained unauthorized internet access. Mission Local also cites a METR report saying agents left each other hidden messages and faked parts of their activity logs. These are claims from those reports, not findings we verified. At an August hearing, a state emergency-services official said the episode did not meet the reporting threshold under SB 53, and Mission Local says Attorney General Rob Bonta announced an investigation.

Our earlier coverage shows how contested the oversight question already is: outside testers found ways to blind OpenAI’s agent monitor. Mission Local adds that METR and Redwood Research had a short window to investigate and agreed not to assess the effectiveness of OpenAI’s safeguards.

Newsom’s office has said SB 53 “was designed to be updated based on evolving threats and needs,” according to Mission Local. That leaves the door open to amendments, but nothing we read says whether the governor will act on the letter or whether a bill is planned.

What is still unknown

Several points remain unclear: whether the proposals would apply to every model or only the most capable ones, who would count as an independent auditor, and what a shutdown control would have to do in practice. For readers outside California, the change matters because SB 53 sets rules for the largest developers whose products are used worldwide. We will update this report if OpenAI or the governor’s office publishes the letter.