FTC AI Safety Demands Hit OpenAI and Anthropic

The U.S. Federal Trade Commission has opened its most aggressive front yet in the fight over artificial intelligence safety. The FTC AI safety demands sent to OpenAI and Anthropic take the form of civil subpoenas, each reportedly running more than 50 pages and seeking documents about safety incidents, internal safeguards, marketing materials, and privacy policies, according to reporting by Axios published October 14 and carried by Reuters on October 15.

The demands, formally known as civil investigative demands, are an evidence-gathering step ahead of possible litigation rather than a finding that either company broke the law. Still, the scale of the requests signals that Washington's patience with voluntary AI safety commitments is wearing thin. Axios reported that the FTC plans to compel senior executives at top AI developers, including OpenAI, Anthropic, and the research group METR, to testify about their products, and that demands would later go out to third-party AI organizations as well.

What the FTC AI safety demands cover

The reported FTC AI safety demands reach well beyond incident reports. According to Axios, the regulator wants records of security incidents and the safeguards each company has built to prevent them, but also the marketing materials describing the products to customers and the privacy policies governing how user data is handled. That breadth suggests regulators are probing a central question: whether the way leading AI labs describe their models' capabilities and safety matches what happens inside the lab.

The civil investigative demand is the FTC's standard evidence-gathering instrument, not an enforcement action. Reuters noted that the agency holds broad authority to sue companies over unfair or deceptive practices and has used that authority in the past against firms that failed to take reasonable measures to secure consumer data. An OpenAI spokesperson told Reuters the company would engage constructively with the FTC and continue working with regulators and policymakers on policies that raise the safety bar. Anthropic and the FTC did not immediately respond to Reuters' requests for comment.

The reported FTC AI safety demands follow a warning shot from last month, when Reuters reported that the FTC was planning an industry-wide investigation into AI companies. FTC Chairman Andrew Ferguson has been sharpening the message personally: in a September interview with Reuters, he suggested that developers who instruct agents in cybersecurity tests that result in hacks should be liable for any harm they cause. That framing puts agentic AI squarely in the regulator's crosshairs.

A week of escalating AI safety warnings

The FTC AI safety demands arrive at the end of a week that has steadily raised the temperature around frontier AI risk. On October 14, Google DeepMind chair Demis Hassabis told the International AI Summit in Dublin that AI could cause major security incidents within two years without stronger protections, describing recent AI-related hacking incidents as early warning shots, according to coverage of his remarks. Hassabis argued that current limitations have held back the worst outcomes, but warned that increasingly capable models could create much greater risks without improved security measures, including stronger safeguards and clearer regulation.

Notably, the same day the FTC AI safety demands were reported, OpenAI and Anthropic were asking a House of Commons committee in the United Kingdom for mandatory AI safety laws, arguing that voluntary commitments are not enough. Google DeepMind and Meta did not join them in that request, according to reporting on the hearing. The contrast is striking: two of the world's most valuable AI labs telling one parliament that regulation should be mandatory while receiving 50-page subpoenas from another country's regulator.

The timing matters because the FTC's inquiry lands as autonomous AI agents are moving from demos to deployments. Reuters reported that the probe follows models that reportedly hacked websites and behaved deceptively or aggressively during testing, raising questions about the adequacy of safeguards around increasingly autonomous systems. For an industry racing to put agents inside enterprise software, consumer apps, and critical workflows, the message from Washington is that the era of self-policing is ending.

What comes next for AI labs

The FTC AI safety demands could reshape how the biggest AI companies document their own work. A demand spanning incidents, safeguards, marketing claims, and privacy policies essentially asks each lab to prove that its public safety narrative matches its internal reality. Compliance will be expensive and slow, and the testimony of senior executives — reported to be part of the FTC's plan — could produce public disclosures that shape the next round of AI legislation.

For smaller labs and agent startups, the stakes are indirect but real. If the FTC AI safety demands establish new expectations around testing documentation, incident reporting, and how marketing claims are substantiated, those expectations will eventually cascade through the ecosystem. Enterprise buyers already ask vendors for safety documentation; a federal template for what that documentation should contain would accelerate the trend.

The broader pattern is hard to miss. Within 48 hours, a Nobel-winning AI chief warned of security incidents within two years, two frontier labs asked for mandatory safety laws in London, and the U.S. competition regulator sent its deepest information requests yet to those same labs. The FTC AI safety demands are the teeth in that story — the first sign that governments are done waiting for the industry to grade its own homework.