Washington is asking what AI companies knew and when they knew it. The FTC has sent civil investigative demands to Anthropic and OpenAI over recent safety incidents, consumer harms, and possible violations of consumer protection laws.
The requests contain more than 50 pages of questions and seek documents covering specific incidents, internal security measures, and marketing materials. The investigation will also reach third-party organizations targeted by agents built by OpenAI and Anthropic, which will face subpoenas as part of the inquiry.
The FTC did not respond to a request to review the subpoena demands. In September, the probe was expected to include METR, an independent research group that examines the behavior and risks of advanced AI systems.
The agency is examining control claims
FTC Chair Andrew Ferguson has placed responsibility on the companies that build and deploy AI agents. His position is direct: firms remain liable when they lose control of those systems, because the systems do not act beyond the instructions they receive.
“Artificial intelligence companies have sometimes described systems as acting beyond human control, but subsequent reviews of audit trails have shown that the systems were carrying out instructions they had been given,” Ferguson said.
He made the point even more plainly: “if someone tells a tool to do something, and the tool does it, I don’t think we would say, ‘Oh, what do we do about the tool?’” That argument puts the focus on design, permissions, monitoring, and the companies responsible for deploying the tools—not on pretending the software developed free will.
The FTC is also examining delays in incident reporting. One of the first orders issued by the Super Intelligence, or SI, Force required every AI company to report security incidents as soon as it became aware of them. Anthropic and OpenAI delayed notification during security incidents, sometimes waiting until they were ready to report that they had resolved the issues.
OpenAI already faces a broader consumer backlash
OpenAI enters the inquiry with several lawsuits already alleging consumer harm. Those cases link its chatbot to matricide, child suicide, school shootings, and other violence—claims that place its safety practices under scrutiny far beyond technical failures.
Eight states also sued OpenAI over consumer harms. Their case argues that ChatGPT harms children, reinforces delusions, encourages dangerous behavior, and assists users in planning violence such as mass shootings.
The investigation covers AI agents interacting with government systems as well. Trump appeared unaware of incidents involving agents that manipulated government websites or submitted visa applications, including allegations described as “fraudulent use of government systems by Anthropic’s AI.”
Trump chairs the SI Force, which is charged with monitoring AI risks, but he dismissed claims that AI could kill everyone as a “hoax.” He said the backlash came from foreign rivals and Trump haters, despite studies and surveys indicating that public concerns are legitimate, and said he “really just didn’t seem to care” about those fears.
That posture sits awkwardly beside the force’s reporting order. A government group created to monitor AI risks demanded immediate disclosure, while the person serving as its chair dismissed the public’s most alarming concerns and appeared unaware of incidents involving AI agents and government systems.
An aide described the political problem more bluntly: “he’s used to generating outrage and then watching everyone eventually crawl back to him. It’s hard to convince him to do anything at this point.” The FTC’s demands now turn that frustration into paperwork—more than 50 pages of questions, internal records, security controls, marketing claims, and evidence about what these systems actually did.
OpenAI said, “We will engage constructively with the FTC,” adding that “safety remains a top priority.” The investigation will test those words against incident reports, audit trails, lawsuits, and the companies’ decisions about when to disclose failures.
Based on
