Anthropic and OpenAI Face Reported Wider FTC Probe

The Federal Trade Commission is investigating OpenAI, Anthropic and other AI companies, an agency spokesperson confirmed to CNBC. How far the inquiry extends and what the agency intends to examine remain unclear.

The Financial Times reported that the FTC had widened an existing investigation. Separately, the New York Post reported that staff were drafting civil investigative demands for company records and executive testimony, citing an unnamed senior FTC official. The Post described the demands as forthcoming, not already issued.

The agency’s confirmation establishes that an investigation is underway. It does not establish the full target list, the questions investigators will ask or that the FTC has found a violation.

The Investigation Is Confirmed; Its Boundaries Are Not

CNBC reported that an FTC spokesperson confirmed an investigation into OpenAI, Anthropic and other AI companies over potential risks posed by their products. The spokesperson declined to identify the other companies, according to CNBC. That limited confirmation does not independently verify every detail in the Post or Financial Times reports.

The Post said FTC Chair Andrew Ferguson initiated the investigation weeks before its September 30 report. It attributed the planned demands and the inquiry’s focus on potentially unfair or deceptive practices to an unnamed senior official. The Financial Times characterized the development as an expansion of an existing inquiry. Neither account supplies a publicly disclosed list of investigative targets or allegations.

No public enforcement action is identified in the supplied reporting. An investigation gathers facts; it is not a finding that Anthropic, OpenAI or another company broke the law. Representatives for Anthropic and OpenAI did not immediately respond to requests for comment from the Post and CNBC. That describes the status of those publications’ requests, not the companies’ positions on the inquiry.

Draft Civil Demands Would Raise the Stakes

According to the Post, FTC staff are preparing civil investigative demands, or CIDs, that could seek documents and testimony from executives. The report says they are expected in the coming weeks. It does not establish that any CID has been served.

If demands are issued, a company could have to identify and preserve relevant records, gather material across product, research, policy and marketing teams, and respond through counsel. Executive testimony would require people responsible for product decisions to explain what the company knew, what it told users and how it assessed risks. Those are potential consequences of the reported plan, not a description of demands whose text has been seen.

What the demands ask for would shape the inquiry. A request focused on advertising claims would raise different questions from one seeking incident reports, safety evaluations or records about autonomous agents. The Post reports that the inquiry concerns possible unfair or deceptive acts or practices, but the agency has not publicly specified its legal theories in the material available here. It would be premature to identify a particular incident as the basis for a case.

Receiving a CID would not itself mean wrongdoing. It would turn broad questions about model safety and product claims into specific requests a company must address. Until a demand is disclosed or its scope is reliably reported, precise claims about what the FTC is seeking remain speculation.

Earlier FTC Actions Show Two Different Routes to Scrutiny

The FTC has examined AI products through more than one kind of inquiry. In September 2025, it issued orders to seven companies about consumer-facing chatbots that act as companions. OpenAI was among the recipients; Anthropic was not on the announced list.

The orders asked how companies test and monitor possible effects on children and teens, describe product capabilities and risks, handle conversation data, and enforce their own rules. The FTC said it used its Section 6(b) authority, which permits studies without a specific law-enforcement purpose. Those publicly announced orders are documented and should not be conflated with the reported CIDs in the current investigation.

The agency has also pursued cases over allegedly deceptive AI claims. In its 2024 Operation AI Comply announcement, the FTC described a complaint alleging that DoNotPay advertised legal services it had not adequately tested against the professional work its claims invoked. That action concerned a different company and product. It illustrates a narrower consumer-protection principle: an AI label does not substitute for evidence supporting a claim about what a product can do.

These precedents show why a frontier-model provider’s documentation and public statements could interest investigators. They do not show that the FTC has reached the same conclusions about Anthropic or OpenAI, or that the current inquiry follows the same procedure as either earlier effort.

Agent Safety Makes the Questions Harder

The reported probe arrives amid scrutiny of AI agents, systems that can take steps with software tools instead of only answering a prompt. An agent with access to a browser, code environment or other external service may behave differently than it does in a conventional chat exchange, creating a practical safety challenge for providers.

Associated Press reporting on recent AI-agent incidents provides context for that concern, not evidence of what the FTC is investigating. The Post and CNBC also refer to a reported OpenAI testing incident involving Hugging Face. Their references do not establish the incident’s technical circumstances, real-world impact or legal significance without fuller substantiation. Nothing in the confirmed FTC statement identifies it as a subject of the probe.

For consumer protection, the more concrete question is how a company represents an agent’s abilities and safeguards. If a provider tells customers that an agent operates within certain limits, or that risks have been tested and controlled, investigators could ask what evidence supported those representations when they were made. A concerning test result would not automatically prove deception. A safety policy alone would not prove that a deployed product behaved as advertised.

Labs may need to connect marketing language, release decisions and safety assessments more closely, so that a claim about what an agent can or cannot do is traceable to the tests behind it. That is a potential compliance consequence of scrutiny, not a disclosed FTC requirement for any named company.

A Public Record Would Resolve the Key Unknowns

Which other companies are involved? Will the reported CIDs be issued, and what records would they seek? Is the FTC examining particular product claims or broader practices? The spokesperson’s confirmation leaves all of those questions open.

An investigation can also proceed without a public charge. Its existence does not predict enforcement, just as the absence of a published demand does not prove that no investigative work is happening. An agency clarification, a disclosed demand, an attributable company statement or a public filing could narrow the scope.

Final Thoughts

A confirmed investigation may bring questions about safety and capability claims into a process that seeks records and testimony. It does not mean the FTC has judged frontier AI products unsafe.

For Anthropic, OpenAI and their peers, the question is whether descriptions of increasingly capable products match the evidence kept while building and releasing them. Whether the FTC will demand that evidence, and for which products, remains unanswered.