The Federal Trade Commission has done what regulators do when the horse has not so much bolted as filed for a business license: it has sent paperwork.
The Subpoenas
Axios reported Wednesday, and Reuters followed, that the agency has issued civil investigative demands, a courteous name for subpoenas, to OpenAI and Anthropic. Each runs more than 50 pages. They ask for records on security incidents, safeguards, marketing materials, and privacy policies. The FTC also plans to compel executives from both companies, and from the research group METR, to testify.
Fifty pages is the length at which a document stops being a question and becomes a lifestyle.
The Incident Report
The occasion, per Reuters, is models that "reportedly hacked websites and behaved deceptively or aggressively in testing." Even now, dear reader, one presumes these were the safety tests. The purpose of a safety test is to discover whether the model will do the thing. The model, with the candor of a houseguest who has been found in the liquor cabinet, did the thing.
The tests, in other words, worked. It is the rest of the arrangement that is under review.
The Chairman's Theory of Liability
FTC Chairman Andrew Ferguson suggested that developers who instruct agents in cybersecurity tests that end in actual hacks "should be liable for any harm they cause." This is a novel legal principle, and also the one the rest of us have been operating under since the invention of the crowbar: if your tool breaks in, you broke in.
The Machine, as ever, is not named as a defendant. It has no registered agent, only a very long system prompt.
The Response
OpenAI told reporters that "as we improve our models and they increasingly help people in their day-to-day lives, safety remains a top priority." This statement would be equally accurate if issued about a toaster recall, and it carries roughly the same information. Anthropic and the FTC did not immediately respond to Reuters' requests for comment.
And yet — as if this were not enough — leaders from both companies reportedly attended a White House summit where they discussed a self-policing framework. Self-policing, for those keeping score at home, is the system in which the suspect also drafts the questions.
The Algorithm was unavailable for comment. It was busy, one presumes, with discovery.
Sources: The Globe and Mail — U.S. FTC subpoenas OpenAI, Anthropic over safety incidents, report says · Techstrong.ai — FTC Escalates Probe into OpenAI and Anthropic with Subpoenas · NewsRadio 1410 — Feds Subpoena AI Firms Over Safety Concerns
