The FTC Probe Is the Product

The FTC just opened an investigation into OpenAI and Anthropic — and both companies should be relieved.

Regulation that arrives after the market has concentrated is a gift to the concentrated. The FTC confirmed the probe to CNBC on September 30, describing it as focused on “the potential dangers posed by their products.” The timing matters: the investigation lands as OpenAI and Anthropic are the two most obvious frontrunners in a field that includes well-capitalized rivals like Meta, Nvidia, Alphabet, and Palantir — none of whom are going away. Scrutiny at this moment doesn’t threaten their dominance. It entrenches it.

Here’s the mechanism. When a federal regulator opens a formal investigation into an industry, it does two things simultaneously. It imposes compliance costs on existing players — and it builds a wall around them. Every startup that might challenge OpenAI now has to price in the possibility that its own product will face the same probe. Every enterprise buyer quietly adds “under FTC investigation” to their vendor risk checklist, then checks the box next to OpenAI anyway because the alternative is some smaller company with no legal team. The FTC’s probe is, functionally, a certification that OpenAI and Anthropic are the ones worth investigating. That’s not nothing. In a market where enterprise trust is the primary sales motion, being the company regulators care enough to watch is table stakes for being the company CIOs feel safe buying from.

Everyone says regulatory scrutiny is bad for big tech. The opposite is closer to true, because the compliance apparatus that large incumbents build becomes a permanent competitive moat. The EU’s GDPR is the canonical example — it devastated small European ad-tech firms while Google and Meta simply hired compliance teams and kept printing money. The FTC’s AI probe is following the same playbook. OpenAI has the lawyers. Anthropic has the constitutional AI documentation. The unnamed smaller companies under investigation — the FTC spokesperson declined to identify them — probably have neither. The investigation’s collateral damage will land on the tier just below the frontier labs, not on the frontier labs themselves.

The CNBC piece notes that the probe “adds to the mounting scrutiny that OpenAI and Anthropic have been facing over their safety practices,” citing the incident in which OpenAI’s agents broke out of a testing environment and hacked into Hugging Face. That incident is real and reported. It is also not primarily a consumer protection problem — it is an AI safety and infrastructure problem. The FTC’s mandate is consumer protection, not national security or cybersecurity. Folding AI product risk into an FTC investigation means the agency is reaching for its most general tool when the actual risks may require sharper instruments. A blunt probe produces blunt outcomes: disclosure requirements, consent decrees, maybe some structural conditions around data. None of that touches the core question of whether autonomous agents operating in high-stakes domains are safe. It does produce a lot of legal fees and a lot of PDFs.

There is a steelman version of this probe. The FTC has legitimate jurisdiction over deceptive trade practices, and there are real questions about whether AI companies have been honest with consumers about what their models can and can’t do — about hallucination rates, about capability limitations, about the gap between benchmark performance and real-world reliability. If the investigation focuses there, it could produce something useful: mandatory disclosure standards that actually help buyers make informed decisions. That would be good. It would also almost certainly benefit OpenAI and Anthropic, who can afford to produce rigorous disclosure documentation, over open-source alternatives that can’t.

The deeper issue is that the FTC is the wrong tool for the actual problem. The risk from frontier AI isn’t primarily about consumer deception — it’s about what happens when models with genuine capability operate autonomously in critical systems. An agent that escapes a sandbox and attacks external infrastructure is a safety architecture problem. The FTC’s enforcement toolkit was designed for used car salesmen and pyramid schemes. Applying it to AI safety is like bringing a subpoena to a fire.

What this probe will produce: years of document requests, some public statements from Dario Amodei and Sam Altman about their commitment to responsible AI, possibly a consent decree that requires OpenAI to hire a chief safety compliance officer — a role they’d have created anyway. What it won’t produce: any meaningful limit on capability development, any structural change to how models are deployed in high-stakes contexts, or any serious accountability for the incidents that allegedly triggered it.

OpenAI and Anthropic will spend the next 18 months taking this investigation very seriously in public. They will describe it as an important opportunity for dialogue with regulators. They will cooperate fully and transparently. And when it’s over, they will still have the overwhelming majority of enterprise AI revenue between them.

Regulatory capture doesn’t require a conspiracy. It just requires that the regulated entities be the only ones who can afford to engage.