FTC OpenAI Anthropic Probe: 50+ Page CIDs on Safety Incidents

The FTC OpenAI Anthropic scoop landed on October 15 via Axios: the Federal Trade Commission sent civil investigative demands to both labs on Wednesday, seeking a wide document haul on safety incidents that have become an industry crisis. Each demand runs more than 50 pages, according to a person familiar with the matter.

What actually happened

Civil investigative demands are the FTC’s pre-litigation evidence tool. They are not a lawsuit, a fine, or a finding. They are a compulsory ask for documents and, later, potentially testimony. Axios reported that the agency plans to compel top AI executives to testify about their products, and that investigative demands will also go to third-party AI organizations at a later date.

The timing sits inside a louder week for AI liability politics. Axios notes that a group of state attorneys general sued OpenAI the same Wednesday under consumer-protection laws, alleging the company misled users — including children — about chatbot safety. Ars Technica’s write-up of the same scoop adds Chair Andrew Ferguson’s earlier comments that firms should be liable when tools follow instructions into harm, and that he rejects the idea of agents as independent actors.

Why the FTC OpenAI Anthropic file matters

The FTC OpenAI Anthropic paper trail, as described, is aimed at consumer harms that could be framed as unfair or deceptive acts: how incidents were handled, what safeguards existed, what marketing claimed, and what privacy policies said. That is a different lane from export controls or model-weight fights. It is also why marketing PDFs and incident timelines suddenly matter as much as eval charts.

Context from Axios: President Trump’s Super Intelligence Force, which Ferguson co-chairs, recently pushed an industry-wide mandate to notify and remediate when incidents occur after what the Force called fraudulent use of government systems by Anthropic’s AI. The Force was set to meet the day after the Axios story. Separately, our Anthropic Scion note covered Anthropic’s internal engineering story — a reminder that these labs are shipping infrastructure at high speed while the compliance stack tries to catch up.

A CID is a document vacuum, not a verdict. Treat “subpoenaed” headlines with that gap intact.

Ars also flags that Reuters floated the probe in September and that METR, an independent research group that investigated recent incidents, could be among third parties subpoenaed. We are not treating that as confirmed service — only as reported context.

What a CID can and cannot prove

A civil investigative demand can compel documents and set up depositions. It does not by itself establish that OpenAI or Anthropic violated the FTC Act. Axios is careful to frame the probe as focused on potential consumer-protection violations, with litigation only a possible later step if the evidence supports it. That gap is where a lot of social media commentary falls apart.

The document categories matter for practitioners. Asking for marketing materials and privacy policies alongside incident records is a signal the agency is testing narrative consistency: what the labs told users versus what internal incident files show. Ferguson’s public line that AI systems follow instructions — rather than “going rogue” — is the philosophical frame Ars highlights. If that frame holds, “the model did it” is a weaker shield.

Parallel private and state litigation already surrounds OpenAI. The FTC track is federal and pre-complaint in these reports. Keep those lanes separate when you summarize the week. Our OpenAI $300 bug bounty archive is a reminder that security theater and security process are not the same beat — and CIDs care about process receipts.

What to do as a reader

If you follow the FTC OpenAI Anthropic story for the paper trail, keep three labels straight: scoop (Axios person familiar), secondary confirmation (Ars/CBS summaries), and primary FTC filing (not public in these reports). Do not invent page counts or quote language that is not in the articles.

Watch for whether third-party CIDs actually go out, whether executives are noticed for testimony, and whether either lab publishes a fuller response beyond OpenAI’s short statement. Compare that regulatory tempo with product news like Shopify Shop MCP when you are mapping how consumer-facing AI surfaces get supervised. Nothing here is legal advice.

Primary sources: the Axios scoop and the Ars Technica report.