FTC Demands Records From OpenAI and Anthropic and Plans to Call Executives

  • AI
  • October 16, 2026
  • 0 Comments

The Federal Trade Commission sent civil investigative demands of more than 50 pages each to OpenAI and Anthropic on October 14, seeking records on safety incidents, protective measures, marketing materials and privacy policies.

Axios reported the demands on October 15, and an agency spokesperson confirmed them. People familiar with the matter said the commission also intends to compel testimony from executives at both companies and will issue requests to third-party organizations working on artificial intelligence at a later date. The stated direction of the inquiry is potential violations of consumer protection law.

A civil investigative demand is the opening move in a consumer protection investigation rather than a finding of fault. The agency uses the instrument to gather documents and testimony under a statutory authority that does not require a court order, and the recipient must respond or negotiate the scope. Fifty pages of specifications describe a wide set of categories, and the breadth is itself a signal about how the commission is framing the questions.

The categories line up with the agency’s established theories. Claims made in marketing that cannot be supported by the product’s actual behavior fall under deception. Data practices that differ from what a privacy policy says fall under the same heading. Safety incidents and the measures taken in response go to whether the company’s public statements about its safeguards match what the records show.

Compelling executive testimony raises the stakes for the individuals involved. Documents can be produced by counsel, and testimony from a chief executive or a head of safety puts statements on the record that plaintiffs in other cases can later use. Companies typically negotiate the scope of such appearances, and the negotiation itself can run for months.

The demand on Anthropic arrives alongside the company’s plans for a public listing. A pending regulatory investigation is a disclosure item in a securities offering, and the company will have to describe it to prospective investors. Its substance matters less for the offering than its resolution date, because buyers of a new listing tend to discount uncertainty they cannot price.

Both companies have signed agreements with the government in other contexts and have published safety frameworks, which the commission will compare with their behavior. The agency has brought cases against software companies over claims about privacy and security that were broader than the practices behind them, and the analysis in those matters turned on the gap between what the company said and what its own records showed.

Third-party organizations are the next target in sequence. Evaluation groups, research nonprofits and enterprise customers hold material that is not in the companies’ control, and requesting it externally is how an investigation tests whether the internal account is complete. The commission has used the approach in other technology markets to reconstruct behavior from sources outside the subject company.

The timing fits a pattern of overlapping scrutiny. State attorneys general filed claims against one developer the same week, and the agency’s own inquiries into the sector have been running for two years. Companies subject to several proceedings at once face a coordination problem: an answer given in one matter is read in all of them, and the privilege questions multiply.

Analysts said the practical outcome of an investigation at this stage is usually a lengthy document production followed by either a closing letter or a negotiated settlement with conduct terms. The agency has not alleged a violation, and the companies have not been asked to change anything yet.

What the demands establish is that the largest model developers are now regulated entities in practice, subject to the same documentary obligations as banks and health companies when the government asks questions. Whether that status produces enforceable standards depends on what the record shows and on whether the commission pursues a case in court or accepts commitments instead.

Other jurisdictions are running their own inquiries, and their questions overlap. European regulators have examined model providers under competition and data rules, and the United Kingdom has asked about foundation model practices. Answers given to one authority are, in practice, available to the others through the public record and through the companies’ own filings.

Document production of this size has an internal cost that rarely appears in public discussion. Material that is responsive to a broad specification must be collected, reviewed for privilege and produced under a schedule, and the engineers who understand the systems involved are the same people pulled into interviews. For a company of a few thousand employees, the request itself reshapes the calendar.

The agency’s own history with emerging technology suggests the process will be slow. Investigations into social platforms and advertising markets have run for years before producing a case, and the remedies that follow are usually conduct commitments negotiated rather than penalties imposed. Companies budget for the legal cost and keep operating.

Related Posts

  • October 16, 2026
  • 15 views
Amazon’s Bee Adds Two Personal AI Devices Starting at $75

Amazon’s Bee released two wearable AI devices on October 15, priced at $74.99 and $99.99, the first update to the product line since Amazon acquired the company in July 2025.…

  • October 16, 2026
  • 11 views
Disney Will Pay $60 Million a Year for 66 Character.AI Employees

Disney plans to pay $60 million a year in salaries to 66 employees hired from Character.AI, and has set aside a $30 million retention pool. Over three years the compensation…