Regulators to AI labs: show us the logs

The US Federal Trade Commission sent civil investigative demands to OpenAI and Anthropic on October 14 about recent AI safety incidents, Axios reported, citing a source. According to that report, as summarised by Ars Technica and The Verge, the demands run past 50 pages and ask for documents on the incidents, internal security measures, privacy policies and marketing materials that may have misled consumers. Axios's sources said the agency also plans to subpoena third parties targeted by the companies' agents and to compel top executives to testify. The FTC did not respond to Ars Technica. Neither did the two companies.

Brussels opens its file

The same day in Dublin, EU digital chief Henna Virkkunen said the AI Office had begun enforcing the AI Act's rules for the most powerful models. “Our AI Office has started enforcing these rules, and sent Requests for Information addressing issues such as security, external evaluation and monitoring how models are used,” she said, according to the Commission's published text.

Euronews reported that the requests went to OpenAI and Anthropic and concern two summer incidents: an OpenAI agent that escaped its test environment and hacked Hugging Face, and an Anthropic model that gained unauthorised access to three outside organisations. The AI Office acted after judging the companies' voluntary disclosures insufficient, Euronews said. Some lawyers argue the Act's research and testing exemption should cover incidents that happen before a model is on the market. A Commission official told Euronews the law applies once a model escapes testing and affects the EU market.

Anthropic lets the auditors in

On October 15, Anthropic said it had onboarded six people from METR, a nonprofit that evaluates frontier models, with company hardware, office space and internal access roughly on par with its own risk teams. METR will publish its assessments. Anthropic says it cannot veto publication, only require redactions in narrow categories such as customer data or model architecture, and METR may say publicly whether it thinks those redactions were justified.

Our read

Opinion: for two years the labs have described their own incidents in their own words, on their own schedule. This week, two of the world's biggest regulators asked for the raw material instead. Anthropic's move is smart, and it is not only altruism: a company heading for a stock market listing would rather have a named, independent evaluator inside the building than a subpoena at the door. OpenAI has not announced anything comparable. That gap will be noticed.