The FTC impersonation rule and AI brand fakes
The Government and Business Impersonation Rule covers firms and agencies. Which AI fakes that leaves actionable today, and which fall to state law instead.
A synthetic ad wearing a real bank's name and logo sits under a named federal rule in the United States. A synthetic video of that same bank's chief executive does not. The gap between those two cases is not a principle anyone argued for. It is an artefact of where a rulemaking stopped.
That distinction decides how fast anyone can act when a fake surfaces, and it decides which team you escalate to. It is worth knowing before you need it, rather than at 9pm on the day a cloned spokesperson clip starts circulating.
General information, not legal advice. Specific incidents need counsel.
What the Impersonation Rule actually covers
The FTC's Government and Business Impersonation Rule took effect in April 2024. Its scope is in the name: impersonation of government agencies and businesses. A fake support line claiming to be a named retailer, a phishing page dressed as a tax authority, a synthetic promotional video using a company's marks and voice as though the company made it — that shape of conduct is what the rule addresses.
The practical value of having a rule rather than only a general deception theory is that the conduct is named up front. Nobody has to argue from first principles that pretending to be a named company is the kind of thing that misleads consumers. It is written down.
A supplemental rulemaking was proposed to extend the same treatment to impersonation of individuals. As of August 2026 that extension appears to remain unfinalised, and I have not been able to confirm a final rule. Treat it as pending rather than as law, and check the current status before you rely on either answer. If it does finalise, it changes the answer to the second half of this post considerably.
The individual gap, and what currently fills it
There is no federal right of publicity in the United States. What exists instead is roughly fifty state regimes that diverge on duration, on who inherits, on what counts as commercial use, and on how much First Amendment breathing room expressive uses get. That patchwork is what a synthetic depiction of a named private person or executive lands in.
Three state instruments do most of the work right now.
Tennessee's ELVIS Act, effective 1 July 2024, added voice as a protected personal right and made unauthorised digital replicas actionable, including against tool providers. That last clause is unusual and is the reason the statute gets cited far outside Tennessee.
California's AB 1836 and AB 2602, both effective 1 January 2025, cover the two ends of the consent problem: replicas of deceased personalities require estate consent, and replica clauses in living performers' contracts are unenforceable without a specific use description and professional representation.
The TAKE IT DOWN Act is the one genuinely federal piece, and it is narrow by design. It criminalises non-consensual intimate imagery including AI-generated material, and imposes a platform notice-and-removal duty within 48 hours. FTC enforcement of the platform obligations began 19 May 2026. It is not a general-purpose likeness statute and should not be described as one.
The NO FAKES Act, which would create a federal voice and likeness digital-replica right, advanced out of the Senate Judiciary Committee on 18 June 2026 as S.4591. It is a bill. It is not law. Plenty of vendor decks describe it as though it were, and that error has been quietly propagating for over a year.
There is also a separate federal hook that catches a lot of individual-shaped fakes without touching publicity law at all. The Rule on the Use of Consumer Reviews and Testimonials, effective 21 October 2024, prohibits AI-generated fake reviews and fabricated testimonials. A synthetic customer who never existed praising a product is squarely inside it, and no likeness right needs to be involved for that to be a problem.
What is actionable today
| The fake | Where it lands now |
|---|---|
| Synthetic ad using a real company's name, logo or marks | Government and Business Impersonation Rule, plus ordinary trade mark exposure |
| Synthetic notice or message styled as a government agency | Same rule |
| Fabricated customer testimonial or AI-generated review | Rule on the Use of Consumer Reviews and Testimonials |
| Synthetic endorsement by a named real person | State right-of-publicity regimes; Section 5 where an implied claim misleads. No finalised federal impersonation rule for individuals |
| Cloned voice of a real person | State law, with Tennessee the most explicit since 1 July 2024. NO FAKES remains a bill |
| Replica of a deceased personality | Estate consent under California AB 1836 and comparable state regimes |
| Non-consensual intimate imagery, including synthetic | TAKE IT DOWN Act; 48-hour platform removal duty, FTC enforcement from 19 May 2026 |
The row that surprises people is the fourth one. A synthetic celebrity endorsement is often treated internally as the most obviously illegal thing on the list, and federally it is the least directly addressed. It is still very much actionable — it is just actionable through fifty doors rather than one.
The trade mark angle is underrated
If a fake uses your marks, you have a trade mark claim regardless of how the pixels were produced, and that route is frequently faster than arguing about synthetic media at all.
This is roughly what Getty was left holding in England. In its case against Stability AI, decided in the High Court on 4 November 2025, Getty dropped the training and output claims mid-trial for want of UK territorial evidence, and the secondary infringement claim failed. What survived was an "extremely limited" trade mark win concerning watermarks. UK training legality was never decided. It is a thin result to build a strategy on, but it does illustrate the pattern: the mark claim was the one that stood up when the novel theories did not.
The mirror image matters for anyone generating. Marks that surface uninvited in outputs are a live exposure, which is the practical reason to run trademark-safe mark development rather than assume a model will avoid producing something registered.
What to do before you generate, not after
Four things worth fixing in process rather than in incident response.
- Separate "depicts a company" from "depicts a person" in your brief. They route to different law and different counsel. A brief that says "a spokesperson for a well-known bank" has quietly combined both.
- Treat named real individuals as a clearance step, not a prompt detail. The clearance path for a real person is documented, slow, and worth starting before storyboarding. Our write-up on how digital replica rights are becoming property covers why that path keeps getting more formal, and the voice and likeness overview covers the voice half specifically.
- Never fabricate a testimonial, even as placeholder. Placeholder copy ships more often than anyone plans for, and the reviews rule does not care about your intent at the point the asset was drafted.
- Log the jurisdictions where the asset will run. Denmark has been reported to have introduced a likeness right covering every person rather than only public figures, though the status of that reporting deserves care and should be confirmed against the official text before anyone relies on it. Territory is not a detail here; it is the first question.
The general hygiene sits in our brand safety checklist for AI content, and if you are building a synthetic presenter as a repeatable asset, the pros and cons of an AI spokesperson covers the disclosure obligations that ride alongside all of this. Endorsement mechanics have their own rulebook, walked through in our guide to the FTC rules for AI creator sponsorships.
FAQ
Does the Impersonation Rule cover a parody of a company?
The rule is aimed at deceptive impersonation. Whether a given parody is deceptive turns on whether a reasonable consumer would take it as coming from or endorsed by the company, and that is a fact question rather than a category exemption. "It's obviously a joke to us" is not the test, because the audience deciding is not you.
If NO FAKES passes, does that replace state likeness law?
Unknown, and the answer depends on preemption language in whatever version ultimately passes, if any version does. It advanced out of Senate Judiciary on 18 June 2026 and nothing about that guarantees a floor vote. Plan around current state law and treat federal coverage as an upside case.
Our fake is on a platform. Do we go to the FTC?
Usually start with the platform. Platform policies on realistic synthetic depictions and impersonation are contractual, enforced by their own teams, and generally faster than any regulator. Regulatory routes matter for scale, repetition, and cases where the platform will not act. Run both when the harm is material.
Is a "not quite" likeness safe?
Recognisability is the thing being regulated, not fidelity. A depiction the audience identifies as a specific person carries the exposure of a replica whether or not the face is exact, and near-miss imitations are the shape several of these statutes were drafted to catch. Style and soundalike imitation that copies no protected work and identifies no specific person remains a different and largely lawful case in the US.