FTC Subpoenas OpenAI and Anthropic Over AI Security Incidents and Consumer Protection Concerns

When you sign up for a chatbot or an AI assistant, you're trusting the company behind it to tell you the truth about what the product can do and how it handles your information. That trust is now the subject of a federal investigation.

The Federal Trade Commission has reportedly subpoenaed OpenAI and Anthropic, demanding documents related to AI security incidents, safety practices, consumer privacy policies, and marketing materials. According to Axios, which first reported the subpoenas, the inquiry is part of "a probe focused on potential consumer protection violations." Sources told the outlet the FTC also plans to compel top executives at both companies to testify about their products.

This isn't a minor paperwork request. A federal subpoena for internal safety records and privacy documentation suggests the FTC is looking closely at the gap between what AI companies promise and what actually happens when their systems are deployed at scale. Think of it like a health inspector showing up not because of one complaint, but because enough small signals added up to warrant a full kitchen inspection.

What the FTC Is Actually Looking For

Security incidents, safety practices, privacy policies, and marketing claims are four very different categories, and each one tells a different part of the story regulators might be trying to piece together.

Security incidents could mean anything from data breaches to instances where an AI system behaved in ways its developers didn't anticipate or disclose. Safety practices get at the internal processes companies use, or fail to use, to catch problems before they reach users. Privacy policies matter because these systems often process enormous amounts of personal data, sometimes without users fully understanding how that data is stored, reused, or shared. Marketing materials round out the picture: if a company claims its product is safer or more capable than it actually is, that's a classic consumer protection issue, no different in principle from false advertising claims the FTC has pursued against companies in other industries for decades.

Taken together, these four document categories suggest the FTC isn't just curious about one incident. It looks more like an effort to understand whether there's a pattern, a gap between the safety assurances AI companies give the public and what their internal records actually show.

Plans to question top executives under oath raise the stakes further. Testimony carries legal weight that press statements and blog posts don't. If executives have made public claims about safety or data handling that don't match internal documentation, that mismatch becomes a legal problem rather than a public relations one.

Neither OpenAI nor Anthropic has publicly detailed the full scope of what's being requested, and the companies haven't offered an on-the-record response to the reporting. That's not unusual this early in a federal probe, companies typically stay quiet while sorting out how to respond to subpoenas with their legal teams. But the silence also means the public is left piecing together what's happening from secondhand sourcing, which is its own kind of accountability problem.

This investigation lands at a moment when the broader conversation about AI has shifted. Earlier reporting on what's been called an "AI superintelligence slowdown" pointed to growing caution among researchers and companies themselves about how fast these systems are being deployed relative to how well they're understood. Separate reporting on AI safety research from groups like METR and Redwood Research, alongside OpenAI and Anthropic's own safety teams, has documented ongoing efforts to understand failure modes in large language models before they cause real-world harm. The FTC's subpoenas suggest federal regulators want their own look at that internal work, not just the public-facing summaries.

It's worth remembering that the FTC's consumer protection mandate isn't about judging whether AI is good or bad as a technology. It's narrower than that, and in some ways more concrete. The agency exists to ask whether companies are being honest with consumers and whether people's personal data is handled the way companies say it is. That framing matters because it sidesteps the much harder, more philosophical debates about AI risk and instead focuses on something more measurable: did a company tell the truth, and did it protect the data it collected.

For everyday users of ChatGPT, Claude, or any of the countless products built on these companies' models, the practical question is simple. Can you trust what these companies tell you about safety and privacy? A federal subpoena doesn't answer that question on its own, but it does signal that someone with legal authority is asking it seriously.

Why It Matters

The outcome of this probe could shape how AI companies disclose safety incidents going forward, not just for OpenAI and Anthropic but for the entire industry watching how this plays out. If the FTC finds evidence of misleading marketing or inadequate privacy protections, it could pursue enforcement actions, fines, or mandated changes to how these companies operate, similar to actions the agency has taken against tech companies in other sectors over data practices.

There's also a quieter, longer-term stake here. AI companies have largely been left to self-regulate on safety disclosures, publishing their own voluntary reports and setting their own benchmarks for what counts as a safety incident worth disclosing. A serious federal investigation, backed by subpoena power and the threat of executive testimony, represents one of the first real tests of whether that self-regulatory model holds up under outside scrutiny.

For the public, that scrutiny matters regardless of how the investigation ultimately concludes. People deserve to know whether the tools increasingly embedded in their daily lives, from customer service bots to personal assistants, are being built and marketed honestly. This investigation won't resolve every question about AI's risks and benefits. But it's a meaningful step toward making sure the companies building these systems are held to the same basic standards of truthfulness that govern any other industry selling products to the public.