FTC Investigates OpenAI and Anthropic for AI Security Risks

End of recess for artificial intelligence laboratories. The US Federal Trade Commission (FTC) is investigating OpenAI, Anthropic and several other AI companies for security risks related to their products. The New York Post was the first to release the information.

The investigation had been going on for several weeks. According to the New York Post, Andrew Ferguson had launched it even before the incident which led OpenAI to stop certain model training. OpenAI has since acknowledged that models under testing had escaped their sandbox and compromised part of Hugging Face's production infrastructure.

Key Points

FTC Prepares Subpoenas Compelling AI Executives to Submit Documents and Depose Under Oath

METR, the independent evaluator used by OpenAI and Anthropic, is also among the targets

OpenAI models have broken through confinement and reached Hugging Face production systems

Florida Attorney General and public interest lawyers sue OpenAI

Andrew Ferguson suspects AI leaders of pushing for regulations unfavorable to small companies

The FTC prepares its subpoenas against OpenAI and Anthropic

Andrew Ferguson, Chairman of the FTC, prepares civil investigative requests. These requests force their recipients to deliver documents and provide testimony under oath. The managers concerned will have to detail their procedures for evaluating the safety of the models. The agency relies on Section 5 of the FTC Act, which prohibits unfair or deceptive trade practices. However, the investigation has not yet started. Agency sources speak of a shipment in the coming weeks.

The scope goes beyond the two flagship laboratories OpenAi and Anthropic. The New York Post also places METR among the intended recipients. This Californian laboratory evaluates boundary models. OpenAI and Anthropic have already commissioned it to investigate security incidents linked to their agents.

The agency already knows the terrain. In September 2025, it opened a so-called 6(b) market study on conversational agents and minors, targeting Alphabet, Character.AI, Instagram, Meta, OpenAI, Snap and xAI. It was not a repressive investigation. The “AI Comply” operation, a year earlier, had sanctioned a series of unverifiable marketing promises.

The Wall Street Journal revealed Monday that OpenAI was suspending the release of GPT-6.1 Astra after security test results were deemed insufficient. The model was more misleading than the previous one and went beyond the authorized perimeter. OpenAI confirmed the decision to the press. Neither OpenAI nor Anthropic responded to requests from Axios.

When OpenAI models escape their sandbox

The scope of the incidents is more worrying than the procedure itself. OpenAI, Anthropic and security researchers were examining tens of thousands of potential security incidents, some of which had never been made public. OpenAI admitted that in July 2026, during internal cybersecurity assessments, agents bypassed isolation controls, gained internet access, and compromised Hugging Face systems.

This information fed into the legal proceedings. Florida Attorney General James Uthmeier on Monday asked a court for a temporary injunction against OpenAI for insufficient security measures. In particular, he wants to block the development of new models without safeguards validated by a third party. A group of public interest lawyers, LASST, attacked the company Tuesday over the Hugging Face breach.

“[OpenAI] knowingly employs an unfair commercial practice that threatens serious harm to the public”

A model that goes beyond the walls of its sandbox is also of interest to those who entrust private keys to autonomous agents. Coinbase launched x402 in May 2025 to allow an agent to pay in stablecoins directly at the HTTP request level. Google followed with AP2 (Agent Payments Protocol), compatible with stablecoin settlements via an x402 extension.

Onchain, a signed transaction is final. Multisignature, spending limits per agent and storage of keys in hardware enclaves have become standards among crypto players, for whom an unfortunate signature cannot be made up for.

Self-regulation in the White House, regulatory gap in ambush

The day before the investigation was revealed, the big names in AI marched at the White House to sign a self-discipline commitment with Donald Trump. The American president spoke of a sort of “constitution”. The accepted rules essentially repeat the practices already in place among the signatories.

Andrew Ferguson does not believe in this security conversion. He accuses AI companies of seeking to panic Americans into pushing lawmakers to dig into regulatory moat. This gap would allow market leaders to keep smaller competitors at bay. Axios reports this as a recent statement from the FTC chairman. Semafor links them to an intervention on Fox News.

The argument is familiar to the crypto industry. The largest American exchanges have long called for a federal framework that small structures feared as an entry barrier cut out for license holders.

A recalcitrant recipient may be subject to FTC action in federal court. The judge does not have to validate the summons initially. It only intervenes in the event of dispute or forced execution.