Whose Face, Whose Voice?

Four things in writing, and no proof

A court will not show a jury a photograph until someone vouches for it. A witness must say it shows what it claims to show, or someone must prove that the camera and the handling of its pictures can be trusted. A website that gets a request to take down an intimate photo of a person needs far less. Under the federal TAKE IT DOWN Act, a request with four things in writing starts a 48-hour clock. None of the four is proof that the photo is real, or fake, or anything else.

The difference follows from what each law is for. A synthetic image or voice clip of a person runs into several bodies of law, and each asks for a different kind of showing because each answers a different question. The reading walks through the act’s removal section, section 3, one part at a time. Each part opens onto another law that reaches the same file. One protects your face, one decides whether a file is genuine, and one decides whether anyone owns it. In the activity you test a harmless file of your own against those laws, then write a removal request for a client, so the statute’s parts become your checklist.

An identifiable individual

The act puts its removal duty on a covered platform, a public site or app built mainly for what its users post. The right to ask belongs to an “identifiable individual.” That is a person who appears in the image and whose face, likeness, or other distinguishing feature, such as a birthmark, is displayed in it. Florida’s own version, in section 836.13, speaks of an “identifiable person” in almost the same words. Both laws reach only intimate images. For any other use of your face, you need a different law.

Florida has had one since 1967. Section 540.08 (opens in a new tab) bars using a person’s name, photograph, or other likeness “for purposes of trade or for any commercial or advertising purpose” without consent. That is the right of publicity, and its reach turns on the word “commercial.” Warner Bros. made The Perfect Storm, a film about a fishing boat lost at sea in a 1991 storm. It invented scenes about the boat’s real captain and briefly portrayed relatives of the crew, and Warner Bros. asked none of them for permission. The captain’s children and others sued under section 540.08. In Tyne v. Time Warner Entertainment Co. (opens in a new tab), the Florida Supreme Court held that “commercial purpose” does not reach a publication, a film included, that does not directly promote a product or service. A broader reading, the court added, would raise what it called a fundamental constitutional concern, because a film made for profit is protected speech just as a book is.

So the statute protects your face against an ad and not against a story, even an invented one. A stranger who puts your altered photo in an ad for a gym has used it for an advertising purpose. The same stranger who puts it in a parody video probably has not, under Tyne.

A federal bill would change that. The NO FAKES Act of 2026 would give every person the right to authorize any digital replica of their voice or visual likeness, meaning a newly created, computer-generated, highly realistic version of them. The right would last at least ten years past death. It steps aside for news, documentary, commentary, satire, and parody, but not where a replica shows sexually explicit conduct. As reported to the Senate, it would also displace some state claims over replicas while leaving alone state laws aimed at sexual or election deepfakes. Whether your own file would count is a real question. The bill reaches a replica in which the person never actually appeared, or a version of a real image or recording whose “fundamental character” has been materially altered. A new background may leave the character of your appearance intact. A cloned voice reading words you never spoke is a performance you never gave.

The depiction, real or forged

The act’s second element asks the request to identify the image and give the platform enough to find it. It never asks whether the image is real. The act covers authentic intimate images and “digital forgeries,” which it defines as images made or altered with software or AI that a reasonable person would find indistinguishable from the real thing. The platform’s duty is the same for both. A platform never has to decide whether an image is genuine. A court sometimes does.

In court, the question is authentication: is the exhibit what the party offering it says it is? Federal Rule of Evidence 901 (opens in a new tab) and Florida’s section 90.901 set the same low bar, evidence sufficient to support that finding, and the jury makes the final call. For a photo or video, Florida courts recognize two routes. One is a witness who knows the scene and says the image shows it accurately. The other is proof about the process that made the image. In Wagner v. State (opens in a new tab), a sheriff’s investigator had hidden a video camera in a car, and the informant who appeared on the tape never testified. The First District held that a recording like that can come in as a “silent witness” once the judge finds it reliable. The judge weighs five things. The first three are the recording’s time and date, any sign of editing or tampering, and the condition of the equipment. The last two are how the equipment and recording were run and kept secure, and testimony identifying the people in it.

A recorded voice works the same way. In Gomez-Martinez v. State (opens in a new tab), the Third District upheld the admission of recorded phone calls although no one who took part in them testified. A detective recognized both voices, the recordings reached him as soon as the calls ended, and the men later did what the calls had arranged.

Now the trouble. A cloned voice is built to fool exactly the witness the first route relies on. A friend who knows your voice could say, honestly, that the clip is you. The federal Advisory Committee on Evidence Rules has studied AI evidence since 2023, and it calls deepfakes easy to make and hard to detect. It has drafted a Rule 901(c) to keep in reserve until it decides a rule is needed. Under the draft, an opponent who claims an exhibit was faked must first offer evidence that could support that finding. Only then must the party offering it prove it more likely than not authentic. The first step keeps “that could be a deepfake” from becoming an objection to every recording. The committee has waited because few courts have met the problem: in a Federal Judicial Center survey of federal judges, only fifteen said they had dealt with deepfake issues.

Do not confuse the draft with proposed Rule 707, the committee’s other AI project. Evidence produced by artificial intelligence is output everyone agrees came from a machine, such as a program’s comparison of two paintings, offered with no expert to explain how the program works. Rule 707 would hold that output to the reliability test for expert testimony. A deepfake raises the opposite problem, because the party offering it does not admit that a machine made it.

A good-faith statement, and no copyright

The third element is a short statement that the person has a good-faith belief the image is not consensual, with whatever facts help the platform see that. The fourth is contact information, and the first is a signature. Anyone who has read the Copyright Act will find the list familiar. Since 1998, section 512 (opens in a new tab) of that act has let a copyright owner have infringing material taken down with a written notice. It asks for a signature, the work infringed, the material and where to find it, contact information, and a good-faith statement. It also asks for a statement, under penalty of perjury, that the sender is authorized to act for the owner.

Why not just use that notice? Because it has to come from the copyright’s owner, and the person in a fake rarely owns anything in it. Copyright belongs to the author. In Thaler v. Perlmutter (opens in a new tab), the D.C. Circuit held that the Copyright Act requires a human author. The Copyright Office treats material an AI system generates, as opposed to what a person made or creatively shaped, as outside copyright, and its registration guidance (opens in a new tab) tells an applicant to disclaim it. So human authorship cuts against the person pictured twice. A fake made by typing a prompt may have no author at all, and the person pictured is not its author either. That person might own the copyright in a photo they took of themselves, if the faker copied it. That claim protects the photo, not the face.

The act follows section 512 in one place and not in another. The act’s shield for a platform that removes material in good faith tracks section 512(g)(1) almost word for word. But section 512 attaches conditions to its shield. The platform must tell the person who posted the material, and if that person files a counter-notice and the sender does not sue, the platform must restore the material to keep its protection. And section 512(f) makes anyone who knowingly and materially misrepresents that material infringes liable for the damage. The TAKE IT DOWN Act has neither. The NO FAKES bill has both: a counter-notice for the uploader and penalties for a false or deceptive notice.

That is a design choice. A person whose intimate image is spreading gets it taken down fast, on their own word, and the act gives the person who posted it no way to answer. Whether that is the right trade is yours to judge.

Forty-eight hours, and after

A valid request, sent through the platform’s own process, requires the platform to act “as soon as possible, but not later than 48 hours” after receiving it. It must remove the image and make reasonable efforts to find and remove known identical copies. The Federal Trade Commission enforces this notice and removal duty. It treats a failure to reasonably comply as an unfair or deceptive practice and can seek a civil penalty for each violation. A person whose request goes unanswered can report the platform on the FTC’s reporting site (opens in a new tab).

Florida set its own deadline first. Section 836.13 (opens in a new tab), amended in 2025, required covered platforms to have a removal process for an altered sexual depiction by December 31, 2025, almost five months before the federal duty took hold. It asks for the same four elements and allows the same 48 hours, and it treats a platform’s failure as a violation of Florida’s Deceptive and Unfair Trade Practices Act. For an image made with AI, a request written to meet one law meets the other.

Removal is not the whole remedy. In Florida, creating an altered sexual depiction without consent, or spreading one maliciously, is a felony. An aggrieved person can also sue for an injunction, $10,000 or actual damages (whichever is greater), and attorney fees. Federal law makes it a crime too, in defined circumstances, to knowingly publish a digital forgery.

In the activity, the middle sections of this reading become your worksheet. It asks how a court would authenticate your file, what you could register, and whether section 540.08 or the NO FAKES bill would protect your face. Then the four elements become a removal request for a client, with a note telling her what the 48 hours mean and what Florida offers beyond them. Start by making the file.