FTC Subpoenas Put AI Safety Claims Under Scrutiny
Matthew Leo · Published October 23, 2026 · News
The United States Federal Trade Commission has reportedly sent civil subpoenas to OpenAI and Anthropic seeking records about safety incidents, security measures, privacy policies and marketing claims. If confirmed, the requests would move the agency’s AI inquiry from public warnings into compulsory evidence gathering.
The central fact still requires caution. Reuters reported the subpoenas on October 15 after Axios cited a person familiar with the matter. The FTC, OpenAI and Anthropic had not confirmed the report when Reuters published. The demands are said to exceed 50 pages, but the documents are not public.
What an FTC inquiry could test
The FTC is not an all-purpose AI regulator. Its strongest route is Section 5 of the FTC Act, which prohibits unfair or deceptive practices. In practical terms, investigators could compare what companies told customers about safety, privacy and security with what internal incident records showed at the time.
That distinction matters. A model producing a harmful or deceptive response in a laboratory test does not automatically prove a consumer-law violation. The harder questions are whether a company made a material claim, possessed contrary evidence, failed to take reasonable precautions or exposed consumers to substantial harm that they could not reasonably avoid.
The reported document categories suggest the agency may be looking across several parts of that chain: incidents, safeguards, data handling and marketing. Reuters also said the FTC plans to compel testimony from executives at major AI developers and the research organization METR.
The agency has already shown its theory
In July, the FTC sought public comment on a proposed policy statement about AI accuracy. It said businesses could violate Section 5 when their representations about an AI system’s objectivity, accuracy or suitability conflict with how the system actually behaves. Earlier FTC guidance also warned AI providers to honour privacy and confidentiality commitments rather than quietly widening data use through changed terms.
That background does not prove the reported subpoenas or predict an enforcement case. It does show why safety reports, red-team results and product advertising may be examined together rather than as separate issues.
Why Canadian buyers should pay attention
An American investigation has no direct authority over Canadian procurement. Still, its evidence could reveal gaps between published safeguards and internal incident handling at services used by Canadian businesses, schools and governments.
Organizations should not wait for a regulator to establish that gap. Contracts should require timely incident notice, clear data-use limits, access to meaningful safety documentation and an exit path when a provider materially changes its service. Mapletechie’s AI-agent approval checklist and its reporting on Anthropic’s false police-tip test show why external actions and incident records need separate controls.
For now, the subpoenas should be described as reported, not confirmed. The next useful evidence would be an FTC filing, an agency statement or a public response from either company.
Sources
- Reuters report on the subpoenas
- Axios reporting on the FTC investigation
- FTC statement on AI accuracy and Section 5
Tags: OpenAI, Anthropic, AI safety, AI governance, privacy