FTC demands reportedly target OpenAI and Anthropic. That is not a verdict

The Federal Trade Commission sent civil investigative demands to OpenAI and Anthropic on October 14, [Axios reported](https://www.axios.com/2026/10/15/ftc‑openai‑anthropic‑ai‑safety‑investigation), citing a person familiar with the matter. Axios described the probe as focused on possible consumer‑protection violations connected to AI safety incidents. According to that report, the demands are more than 50 pages each and seek documents about security incidents, safeguarding practices, marketing and privacy policies. Axios also reported that the agency plans to compel senior executives to testify. Those details come from Axios and its unnamed source. The reporting available for this story does not include the demands themselves or a public FTC complaint against either company. OpenAI told Axios that safety remains a priority and that it would engage constructively with the agency.

The FTC’s own [CID guidance](https://www.ftc.gov/business‑guidance/blog/2025/03/did‑your‑business‑receive‑cid‑ftc‑means‑business) describes a civil investigative demand as legally enforceable compulsory process. It can seek documents, testimony, physical material, reports or answers to questions during an investigation. The agency can send a CID to a person or company it thinks may have violated the law. It can also send one to a third party that may simply hold relevant information. The FTC says the demand should identify the subject of the investigation and the laws or rules that may be involved. That distinction is essential: receiving a demand is not a finding that the recipient broke the law. It is an evidence‑gathering step that can end without a case, support a later settlement or contribute to litigation.

Axios says the FTC is asking for records about model security incidents and the safeguards used to prevent harm. Marketing material and privacy policies are also reportedly within scope. That combination suggests the agency is looking beyond whether a model produced a dangerous result. Consumer‑protection questions can include what a company promised, what it knew internally, how it described protections and how it handled people’s data. This is an inference from the reported document categories, not a statement of the agency’s final theory. The FTC maintains a broad [AI enforcement and policy index](https://www.ftc.gov/industry/technology/artificial‑intelligence), including prior matters involving misleading AI claims and data practices. Those earlier actions show the kinds of legal tools the agency has used; they do not determine the outcome of this reported investigation.

An investigation does not establish that one chatbot is safer than another, and the absence of a public demand would not prove that another system is safe. Regulatory process and product risk are different questions. For people using ChatGPT or Claude, the immediate controls remain practical: do not give a chatbot secrets it does not need, verify consequential claims, keep a human responsible for medical, legal, financial or employment decisions, and review the service’s current data settings. Organizations should preserve incident records and make public claims that match tested controls. A promise such as “safe,” “private” or “secure” can carry more weight than an internal aspiration when customers rely on it.

The investigation becomes more concrete if the FTC publishes a complaint, settlement, closing statement or other official record. Until then, the scope and status may remain partly private because investigative material is often confidential. A future filing would still need to be read carefully: allegations in a complaint are not final findings, while a negotiated order may resolve a matter without the company admitting every allegation. The remedy, factual record and procedural posture matter. Follow company‑specific changes on the [Anthropic organization page](https://opentools.ai/organizations/anthropic). *Figure by OpenTools Team; no third‑party expressive material used. The figure summarizes Axios reporting and FTC process guidance and does not depict a finding of wrongdoing.*