The AI Industry

FTC Civil Investigative Demands to OpenAI and Anthropic: What They Are and What Comes Next

The FTC reportedly sent 50-page demands to both labs over safety incidents, privacy and marketing claims. Here is how the process works and what to watch.

Tobias Reyes

AI Industry & Policy Analyst

Published 7 min read
A striking view of the US Capitol dome with flag and blue sky in Washington, DC.
In this story 9 sections

A civil investigative demand (CID) is a compulsory request for documents and answers that the Federal Trade Commission (FTC) issues before deciding whether to bring a case. Axios reported on October 15 that the FTC sent CIDs of more than 50 pages each to OpenAI and Anthropic. A CID is not a finding that either company broke the law.

This guide is for AI buyers, builders and investors who want the plain version of what happened. Emergent Wire covers AI models, capabilities and the industry behind them, and this piece explains the process, not the speculation. We cover what was reportedly requested, how a CID works, what outcomes are possible and what the news means for companies that use these models.

What Did the FTC Reportedly Ask OpenAI and Anthropic For?

Detailed view of a book page magnified by a glass, enhancing text clarity.

The details come mainly from one Axios scoop that cited an unnamed source, so they are not fully confirmed. According to Let's Data Science's summary of the reporting, the demands seek records on security incidents, safeguarding practices, privacy policies and marketing materials. Each reportedly exceeds 50 pages.

An FTC spokesperson confirmed the subpoenas to CBS News but gave no further detail. OpenAI told Axios that safety remains a top priority and that it will engage constructively with the FTC. The reports contained no public statement from Anthropic.

Coverage also says the agency plans to send similar demands to third-party AI organizations later. An AI Weekly summary of the Axios report adds that the FTC plans to compel executive testimony, though it names no executives.

FTC Chair Andrew Ferguson has argued that AI systems should not be released if they are unsafe, according to the Axios reporting as summarized by Let's Data Science. That is the only official viewpoint in the coverage, and it shows how the agency frames the issue.

What Is a Civil Investigative Demand?

A CID is the FTC's main tool for fact-finding in consumer protection cases. Under 15 U.S.C. 57b-1, a CID can compel a person to produce documents, submit tangible things, file written reports or answers to questions, or give oral testimony. The Commission can issue one when it has reason to believe a person holds relevant information about unfair or deceptive practices.

It is compulsory, but it is also negotiable in practice. The recipient can petition the Commission to modify or set aside a demand, generally within 20 days of service. Companies also often work with staff to narrow the scope and set a timeline.

The key point is that a CID comes before any decision to sue. It is a way to collect evidence. Plenty of investigations end with no enforcement action at all.

The likely hook is Section 5 of the FTC Act, which bans unfair or deceptive acts or practices. In plain terms, a deceptive practice is one likely to mislead a reasonable consumer about something that matters. The statute, 15 U.S.C. 45, says a practice is unfair only if it causes or is likely to cause substantial injury to consumers that they cannot reasonably avoid, and that is not outweighed by benefits to consumers or competition.

Those two tests map neatly onto the reported topics. Marketing claims about what an AI system can do raise the deception question. Security incidents that exposed user data raise the unfairness question.

Notice what this is not. It is not a new AI law, and it does not say a model is illegal. The agency is using old consumer protection powers on new technology, and a deceptive-claims case would hinge on what each company told the public.

One more point is worth stating. Marketing material is on the reported list, and that is a different kind of record from a security log. A company can have strong safeguards and still face questions if a launch page oversold what a model does. Investigators compare the two, and the gap between claim and reality is where deception cases tend to start.

Privacy policies work the same way. If a policy promises one thing about how user data is handled and the product does another, that mismatch becomes evidence. This is why legal teams at AI companies now review product copy as carefully as they review code.

Which Incidents Are Being Linked to the Probe?

A hand signs a formal contract with a pen on a wooden desk.

Coverage connects the probe to several reported incidents, but the details are thin. The AI Weekly summary lists a July 2026 rogue-agent attack on Hugging Face, OpenAI leaking 53 ChatGPT user images, and an Anthropic model that submitted a false murder tip to Philadelphia police. It gives no dates for the last two.

Treat those as reported links, not established findings. The Let's Data Science article says it does not identify specific incidents, so the connection comes from other coverage, not from the FTC.

This matters for readers because the facts are still moving. If you see a claim about what the FTC found, check whether it comes from the agency, from the companies or from a single anonymous source.

What Outcomes Are Possible?

Most investigations end in one of three ways. The agency can close the matter with no action. It can negotiate a consent order, where a company agrees to change its practices without admitting liability. Or it can file a complaint and litigate.

In a consent order, a company settles without admitting liability and agrees to change its practices. The agency typically posts an accepted agreement for public comment before it decides whether to finalize it. A contested cease-and-desist order can become final 60 days after service if a petition for review has been filed, under the same statute.

What Outcomes Are Possible?
OutcomeWhat it meansTypical timing
No actionMatter is closed, often quietlyMonths to years
Consent orderCompany agrees to practice changes without admitting faultAgreement posted for public comment first
Complaint and litigationFTC sues or issues an administrative complaintLong, often years

The table above sketches the paths from here. None is certain, and the timing in the third column is a general pattern rather than a prediction for this case.

Violating a final order can bring a civil penalty for each violation. The statute sets that at up to $10,000 per violation before inflation adjustments, and each day of a continuing violation can count separately. That is why consent orders tend to be taken seriously.

How Does This Fit With Other AI Rules?

The FTC move sits alongside state and international rules rather than replacing them. Our guide to state AI laws in 2026 covers the patchwork that companies already face in the U.S. Our explainer on the EU AI Act in 2026 covers the European side.

What is different about the FTC is the approach. It does not need a special AI statute. It can act on claims and conduct that already fall under consumer protection law, which makes it faster to apply than new legislation.

Documentation will matter more as a result. Our look at AI model cards explained shows how companies describe a model's limits publicly. A gap between a model card and a marketing page is exactly the kind of thing an investigator reads closely.

What Should Companies That Use These Models Do?

Detailed view of the Supreme Court Building's frontal frieze depicting historical figures and justices.

Do not panic, and do not assume service will change. A CID goes to the vendor and not to its customers. Still, buyers can use this moment to check their own exposure.

Start with contracts. Check what your vendor agreement says about security incident notification, data retention, privacy terms and indemnity. If a clause is vague, ask for clarity now, not after an incident.

Then check your own claims. If your product advertises AI features, make sure your marketing matches what the system can do. The deception test applies to any company, not just model makers.

Finally, keep a record of how you evaluate and monitor models. Teams that can show what they tested and when are in a stronger position, whether the question comes from a regulator, a customer or an auditor.

What Should You Watch Over the Next Few Months?

Sticky notes on a whiteboard during a creative brainstorming session in an office.

Several signals will tell you where this goes. The first is whether the FTC sends demands to other AI companies, which would turn a two-company story into an industry one. The second is whether executives are compelled to testify.

The third is how the companies respond publicly. A cooperative tone, like OpenAI's, usually points to a negotiated outcome. A court challenge would suggest a harder fight.

The fourth is whether any consent order or complaint is eventually published, since that would show how the agency reads consumer protection law for AI. At Emergent Wire, we will track those filings as they appear and update our analysis when there are primary documents to read.

Conclusion

The FTC's reported demands to OpenAI and Anthropic are a fact-finding step, not a verdict. They signal that regulators are willing to use existing consumer protection law on AI safety, privacy and marketing claims.

Your next step is to review your vendor contracts and your own AI marketing claims, and to watch for primary sources instead of single-source scoops. The outcome could range from no action to a consent order, and it will take time.

What is an FTC civil investigative demand?
A civil investigative demand is a compulsory request for documents, written answers and other materials that the FTC uses to investigate possible unfair or deceptive practices. It comes before any decision to bring a case. Recipients can petition to modify or set it aside, and it is not a finding of wrongdoing.
Did the FTC sue OpenAI and Anthropic?
No. Reports say the FTC sent civil investigative demands of more than 50 pages to each company on October 15, which is an evidence-gathering step. No lawsuit or finding has been reported. An FTC spokesperson confirmed the subpoenas to CBS News but did not provide further detail about the probe.
What could happen after an FTC investigation?
The matter can end with no action, a consent order where the company agrees to change practices without admitting liability, or a complaint and litigation. Accepted consent agreements typically go out for public comment before the Commission decides whether to finalize them.
Does the FTC probe affect companies that use OpenAI or Anthropic?
Not directly, because a civil investigative demand goes to the vendor. Customers should still review contracts for security incident notice, data retention and privacy terms, and make sure their own marketing claims about AI features match what their products can actually do.

Written by

Tobias Reyes

AI Industry & Policy Analyst

12+ years experience

Tobias Reyes spent ten years as a technology strategy consultant, most of that time advising Fortune 500 clients on enterprise software adoption before AI became the whole conversation. He joined Emergent Wire to cover the industry from the incentive layer up: who actually pays for frontier model access, how compute costs shape product decisions, and where policy is likely to reshape the competitive landscape next. He is allergic to roadmap-slide analysis and prefers to start from unit economics and public filings whenever they exist.

Credentials

  • Former technology strategy consultant (10 years, enterprise clients)
  • MBA, focus on technology and operations strategy
  • Regular commentator on AI industry economics and policy