Deepfake removal in 2026 runs on four kinds of machinery, and picking the wrong one costs you days. An intimate fake falls under the federal TAKE IT DOWN Act, and covered platforms must remove it within 48 hours of a valid request. A fake built from your actual photos, videos, or audio is a copyright case, and the DMCA works. An impersonation, scam, or smear generated from scratch runs through platform synthetic-media and impersonation policies. A fake parked on a hostile site escalates to the host, the registrar, and eventually a court.
What never works is improvising, a copyright claim on a face you don't own, a polite email where a statutory demand belongs. The sections below follow the order you'd act in.
Sorting your fake into the right removal lane
Speed depends on filing at the right door, and the doors look alike only from a distance. Four lanes:
- Intimate fakes. Synthetic nudity, a sex tape that never existed. The TAKE IT DOWN Act applies: 48-hour removal on covered platforms, and publishing it is a federal crime.
- Fakes built from your real material. An edited photo, a manipulated video, your recorded voice spliced in, your song on the soundtrack. That's infringement of work you own, and the DMCA applies.
- Impersonation, fraud, and smear fakes. Generated from scratch but using your face, name, or voice. Platform synthetic-media, impersonation, and fraud policies are the machinery.
- Anything on a rogue site. The escalation lane: host, registrar, search engines, court order.
A single fake can fit two lanes: an intimate fake cut from your real photos is a TAKE IT DOWN case and a copyright case at once. File both in parallel. Nothing about one lane requires abandoning the others.
Why the copyright reflex is usually the wrong door
The first instinct for most people is a DMCA takedown, and it's usually wrong. Copyright protects works of authorship: photos you took, videos you recorded, songs you wrote. Your face, name, and voice are protected as identity, not authorship, and identity lives under different law entirely, publicity rights, privacy torts, impersonation rules. Filing a copyright claim against content you hold no rights in exposes you to damages for knowingly material misrepresentation under 17 U.S.C. § 512(f), and a rejected notice burns the hours that mattered.
The exception is the fake built from your real material. If someone edited your actual photograph, manipulated your real video, spliced in your recorded voice, or set the fake to your song, the underlying work is yours and the notice works like any other copyright takedown, the sequence is in our guide to filing a takedown notice. One distinction matters: built from is not trained to mimic. A voice model trained on your recordings gives you no copyright hook in the output, because the output isn't a copy of your recordings.
And nobody may own the fake itself. Purely AI-generated output can lack human authorship altogether, who owns AI-generated content is unsettled in exactly the way that hurts a copyright theory.
Intimate deepfakes: the TAKE IT DOWN Act's 48-hour clock
The Act, signed in May 2025, closed the deepfake loophole directly: non-consensual intimate imagery explicitly includes depictions created or altered by artificial intelligence, and covered platforms, the Act's term for services that host user-shared content, must remove it within 48 hours of a valid request from the person depicted or their representative. Publishing it is a federal crime, independent of what any platform does.
A valid request asks less than people fear: identify the content, state that you're the person depicted or an authorized representative, state that it's non-consensual. You don't need to identify the poster or prove intent. Platforms must also take reasonable steps, where feasible, to keep the same content from going straight back up.
Keep records of your request and the clock. If a covered platform misses the window, the removal duty is enforced by the Federal Trade Commission. Platform-level intimate-imagery reports are worth filing the same hour too: most major platforms run their own expedited queues, and nothing about one process excludes the other. If the person depicted was a minor, the National Center for Missing & Exploited Children runs a free removal service that predates the Act. For the full victim-side sequence, including fakes built from real intimate photos, see our NCII removal guide.
Impersonation, fraud, and smear fakes: the platform policy lane
When the fake isn't intimate, a political smear, a fake endorsement, a cloned voice selling crypto, platform policy is the fast lane, and the details matter as much as the channel.
Report as the person impersonated. Platforms weight the victim's own report far above third-party flags; it's the single biggest speed factor. Use the impersonation complaint, not the general flag, taking down impersonation accounts covers that path, and it reaches the synthetic version of you as well as the account version.
File under the AI policy too. Every major platform now requires realistic synthetic media to be disclosed, and undisclosed synthetic depictions of real people saying or doing things they never did are removable under manipulated-media rules. The angle that moves reviewers is deception: clearly labeled parody survives most platforms; an unlabeled fake of you saying something you never said doesn't. When a fake endorses a product or scam, add the fraud report, counter-fraud queues move fastest of all.
Two caveats. Platforms sometimes answer with a label instead of a removal; if so, appeal through the privacy complaint. And outside the United States, outcomes often ride on different law, the EU's AI Act transparency requirements, or the Digital Services Act's notice-and-action process, rather than the TAKE IT DOWN Act.
Evidence: proving something is fake that never happened
Documenting a deepfake inverts the usual copyright problem. You're not proving you own the content, you're proving the content lies about you.
Capture first, before reporting anything: full-page captures with the URL visible, the account handle, the date, the view count. Fakes get deleted mid-review like everything else, and reposts later look like isolated incidents unless your records connect them. Save the page around the fake too, captions and comments, because defamation turns on context.
Then build the side-by-side record: your dated originals from the same era, setting, or outfit; your real accounts' posting history; whatever establishes what you actually look and sound like at that moment. Why dated evidence wins arguments applies here exactly as in copyright disputes, and if your fake is the derived kind, you'll additionally be documenting ownership of the underlying photo or video.
The classic tells, mangled hands, warped teeth, text that won't render, audio drifting out of sync, help, but don't let them become your burden of proof. Platforms run their own checks, automated detectors are unreliable in both directions, and nothing requires you to prove a forgery forensically before reporting.
Fakes on hostile sites: hosts, registrars, and the court-order ladder
Some deepfakes live where policies go to die, image boards, forums, offshore sites whose abuse process is a contact form nobody reads. The site owner is the wrong target. Find the site's host and file there instead: hosting providers answer abuse complaints because their own acceptable-use policies forbid the material, and they lose nothing by deleting a stranger's page. Registrars are a weaker lever but a real one, a dedicated impersonation or defamation domain sometimes violates registrar policy outright.
When nothing voluntary works, the difference between a takedown notice and a court order becomes the whole story. An uncooperative host that ignores policy will still honor a court order, and subpoena power is how anonymous operators get named. This lane isn't fast. It's the only one that works against sites with no other reason to care.
After removal: search echoes, mirrors, and the repost problem
Source removal isn't the finish line. Search results keep showing the snippet, caches keep serving the page, mirrors keep the file alive, and the fake that comes down Tuesday reappears Wednesday on a new account. Treat the aftermath as part of the remedy.
Google runs a dedicated removal process for non-consensual and fake explicit imagery, and covers impersonation and doxxing in defined situations; major competitors run comparable policies. Where the fake contains material you do own, your footage, your recording, your music, the standard Google delisting process handles that slice on ordinary copyright grounds. Delisting doesn't delete the source, so it's the second layer, after the removal demand.
Monitoring closes the loop. Reverse image searches on a few key frames, alerts on your name plus the words likely to tag along with a repost. Continuous monitoring closes the gap, Protection Pro watches for reuploads and files fresh reports, so each new copy meets a fresh 48-hour clock. Verified, consistently active official accounts help too: they make impersonation reports faster and cut the credibility a fake borrows by default.
When it's a campaign: the claims that stack
A coordinated deepfake push, the same fake seeded across accounts, forums, and sites, is a legal case as much as a content case, and the two tracks don't wait for each other.
On the legal side, claims tend to stack. Defamation is the core one, and false statements imputing sexual conduct are defamation per se in most states, so damages are presumed. False light, harassment and stalking statutes, and publicity rights for commercial uses layer on, and a federal civil action exists specifically for intimate digital forgeries, 15 U.S.C. § 6851, alongside the Act's criminal provisions. State specifics vary enough that counsel should pick the combination; the takedowns shouldn't wait. Hand the volume of URLs, accounts, and reuploads to our personal takedown service so removals continue while a lawyer works the operator.
One boundary worth checking: if what's being faked is your product rather than your person, fake ads, fake product pages assembled from your creative, that's usually plain copyright infringement of your material, and brand protection monitoring catches lookalike domains and copied creative before a fake-ad campaign scales.
Frequently asked questions
Is a deepfake of me copyright infringement?
Usually not. Copyright covers works of authorship, photographs, recordings, videos, while your face, name, and voice are protected as identity under different law. The exception is a fake built from your actual material: your photo edited, your video manipulated, your audio spliced in. That infringes your copyright. A fabrication from scratch typically infringes no one's, and may have no owner at all.
Do I have to prove the video is fake before a platform will act?
No. A valid TAKE IT DOWN Act request requires identifying the content and stating that you're the person depicted or their representative and that the material is non-consensual, the statute deliberately avoids putting the burden of proof on the victim. For non-intimate fakes, platforms run their own reviews. Strong evidence shortens review time; it isn't a precondition for filing.
What if a platform ignores my TAKE IT DOWN Act request?
Document everything: the request, the date sent, any confirmation, and the moment the 48-hour window closed. The removal duty is enforced by the Federal Trade Commission, so that's where the complaint goes. In parallel, publication of the fake is a federal crime you can report to law enforcement, and the escalation ladder of host, registrar, and search engines still applies.
Can deepfakes be removed from Google search results?
Yes. Google runs a dedicated removal process for non-consensual and fake explicit imagery, plus defined policies covering impersonation and doxxing. Delisting doesn't delete the file at the source, so pair search removals with removal demands to the platform and host. Where the fake also contains material you own, ordinary copyright delisting covers that portion.
Is creating a non-sexual deepfake of someone illegal?
It can be. Non-intimate deepfakes cross into illegality when they defame, constitute harassment or stalking under state law, misuse identity for commercial purposes, or commit fraud. Several states also restrict deceptive political deepfakes near elections. The federal criminal prohibition is narrower, non-consensual intimate imagery, synthetic or otherwise, but the civil claims are broad, and they stack.
What to do next, in order
- Capture everything, the fake, the full page, the URL, the handle, the date, before anything gets deleted.
- Sort the fake into its lane: intimate, built from your material, or impersonation and fraud.
- File the matching report today. Cite the TAKE IT DOWN Act by name if the fake is intimate; report as the impersonated person if it isn't.
- Build the side-by-side record with your dated originals, and log every URL, report, and case number.
- Circle back for the echoes: search results, caches, mirrors, reuploads.
- If it's a coordinated push, get counsel for the operator while a service runs the content volume.
The fake never happened. The machinery for erasing it does.