Deepfake removal and deepfake laws: cases and your options
By
Ceartas Team
A deepfake takes a real person's face or voice and puts it somewhere they never agreed to be: in sexual imagery, in a fabricated statement, or in an advert selling something they have never heard of. It hits creators, presenters, executives and public figures hardest, because their faces are public, trusted and easy to copy, and it hits the people who manage them too.
This page tracks the laws, court cases, police action and platform decisions that change what you can actually do about it. It is updated as things happen, newest first, and every entry links to its source.
The short version
Laws are arriving, but they are local and often not final. A Minnesota ban survived an early challenge in one state, and Denmark's face and voice bill is still a bill.
Criminal cases stall at identification and borders. In the UK, reports have closed because no suspect could be identified, or because the person traced was abroad. Removal does not need a name.
Removal at source works when someone does it. The Manhattan District Attorney took twelve sites offline by seizing their domains, rather than asking a search engine to hide them.
Hidden is not removed. A study traced the ordinary infrastructure companies still serving deepfake abuse sites, which is why a link vanishing from search does not mean the page came down.
Platform reporting is slow, and sometimes says no. Fake ads using real faces have stayed up for weeks, been refused, or needed a social media company's oversight board to order removal.
Nobody tells the person whose face was used. UK police counted £24.3 million lost to AI enabled investment fraud, including fake endorsements, in eight months, and the reports came from the people who lost money.
What you can do today
None of these steps depends on a new law, and none of them guarantees an outcome or a timescale.
Keep evidence before you report. Screenshot the post or advert, and note the URL, the account or advertiser name, the date you found it and where it pointed. Keep a list of locations, not copies of the material itself.
Report it to the platform through the right route. Sexual imagery made without consent has its own reporting route on most platforms, separate from ordinary reports. Paid adverts are handled separately from posts, so report the advert as an advert and the account as impersonation.
If the answer is no, change the basis. A refused policy report is not the only door. The same fake may also be an impersonation problem, a likeness or publicity rights problem, or a breach of advertising rules where it runs.
Send a legal notice to the host. Where copyright, likeness or platform policy gives you a basis, a notice to the site or host that stores the file asks for the file itself to go.
Use search engine removal, and know its limit. Getting a result delisted cuts traffic, but the page stays live for anyone with the link. Track delisting and removal as two separate outcomes.
Go to the police where it is criminal, and start removal the same day. An investigation is about the person who made it, not the copies. In the UK, sexual deepfakes can be reported as a sexual offence.
Search the ad libraries and tell people. Search your name and handle in the platforms' public ad archives, post one plain "this is not me" statement, and warn brand partners before they find it.
Expect it to come back. Removed material returns under new accounts and new domains, so check again on a schedule.
If you would rather not do this alone, see how Ceartas handles deepfake removal.
What changed: explicit and non consensual deepfakes
Newest first.
22 October 2026: A closed UK deepfake case is reopened as a sexual offence
On 29 September an investigative newsroom and a UK newspaper reported that police in England and Wales had charged nobody over an estimated three million sexualised deepfakes (the newsroom's estimate), and women who reported them were told no suspect could be identified. The company behind the AI chatbot used to make the images said in September that it had not been possible, in the time available, to look into communications with South Wales Police. On 22 October the newsroom reported that South Wales Police had reopened one closed case under sexual offences law, after the newsroom passed it new information. No arrest, charge or identification has been reported. Identification can take months; removal does not need to wait for it.
Sources: the October report and the September report.
8 October 2026: Denmark introduces a bill to put faces and voices inside copyright law
Denmark's government was due to introduce a bill amending its copyright law to ban sharing realistic deepfakes of a person's face or voice made without consent. The country's culture minister said a main goal is to get tech companies to take manipulated content down quickly. Satire, parody, caricature and criticism of power and society would be exempt. It is a bill, not law: the minister hopes it takes effect at the start of 2027, and the reporting describes no penalties and no takedown deadline. It applies to Denmark only, and a right still leaves someone to find every copy and ask.
Source: read the report. Read our analysis: what Denmark's face and voice bill would change.
30 September 2026: A study maps the companies still serving deepfake abuse sites
Researchers at a US university and a UK university identified 88 sites actively hosting non consensual AI intimate imagery, after ruling out 312 of an initial 400 as keyword squatting pages. Five infrastructure providers came up repeatedly: a CDN and DNS provider, a large search company, a domain registrar, a website publishing software project and a privacy focused email provider. This maps infrastructure provision, not wrongdoing or intent. The search company said it cannot investigate without specific domains, the publishing software project said it is open source software rather than a host, and the other three did not respond. Why it matters: a page delisted from search can still be fully online, and each layer (registrar, host, delivery network) can do something different about it.
Source: read the report.
17 September 2026: A social media company's oversight board orders deepfakes of two women taken down
The oversight board of a major social media company ruled that AI generated videos putting fabricated statements into the mouths of a Scottish councillor and a health campaigner broke the company's hateful conduct rules, and ordered their removal. It also made nine recommendations, including wider "high risk" AI labels and less algorithmic distribution of deceptive AI material. Those recommendations are not binding, and the company has not adopted them, so the rules have not changed. Why it matters: the videos had been on the company's main social network, and it took a separate body's formal decision to get them removed. Reporting alone is not always enough.
Source: read the report.
14 September 2026: Manhattan prosecutors seize twelve sites selling sexual deepfakes
The Manhattan District Attorney's Office seized the domains of twelve websites selling AI generated sexual images of about 1,200 people who never consented, including actors, politicians, athletes, musicians and influencers. No arrests or charges have been announced; the operators are under investigation. The office says it will keep monitoring to stop the sites operating under new domains. This was removal at source, not a search delisting: the storefronts went, not just the links. It rests on New York law and does not create the same route in the UK or Ireland.
Source: Manhattan District Attorney's Office. Read our analysis: what the Manhattan seizure achieves.
4 September 2026: Minnesota's AI nudification ban survives an AI company's challenge, for now
A federal judge in Minnesota denied an AI company's request for a preliminary injunction against the state's law banning tools that create fake nude images of identifiable people, after denying a temporary restraining order on 31 July 2026. The law, in effect from August, also bars operators and developers from enabling that use and from advertising such tools. This is not a final ruling that the law is valid, the challenge continues, and it binds one US state. A law establishes the wrong; it does not take any image offline.
Source: read the report. Read our analysis: Minnesota's ban after the AI company's challenge.
What changed: deepfake scam endorsements
Newest first.
30 October 2026: UK police put losses to fake AI endorsements at £24.3 million
City of London Police figures, released in late October and covered by an AI news site, put losses to investment fraud using AI, including fake endorsements from public figures and trusted financial institutions, at £24.3 million between January and August 2026. That came from more than 1,300 reports, 1,033 of them from people aged sixty or over. A commander at the force said: "Increasingly, we are seeing various forms of AI, including deepfakes, being used." No one is named and no arrests are reported. Why it matters: fraud reports come from the people who lost money, so the person whose face was used is usually never told.
Source: read the report.
14 October 2026: The Bank of England Governor says platforms carry "no real liability"
Speaking at the IMF annual meetings in Bangkok, the Governor said: "There's no real liability for the provision of the platform to what's on it." He was reported as contrasting platforms with newspapers, which are responsible for what they publish. Separately, in June 2026, AI generated scam adverts using his likeness circulated on a social media platform and pointed people to fraudulent websites, and the Bank said they were very hard to track down and remove. The speech changes no law and nobody has been charged or found liable. If the Bank struggled to get fakes down, removal is the hard part.
Source: read the report.
22 September 2026: A pensioner loses £40,000 to a deepfake on a fake TV debate set
A 72 year old man in Guildford lost £40,000 after an AI generated video, showing a UK politician confronting the Bank of England governor against a fake backdrop of a UK broadcaster's political debate show, led him to a bogus investment platform. None of the people or organisations impersonated had anything to do with it. No police force is cited and no arrests are reported. Why it matters: the fraud ran on borrowed credibility, and the people whose faces and formats were used had no way of knowing it was happening.
Source: read the report.
17 September 2026: A broadcaster says a social network refused to remove a deepfake ad of him
A consumer finance broadcaster says a major social network told him a deepfake ad using his cloned face and voice to sell insurance "meets their standards" and would not be taken down, after weeks of trying. The network's owner has made no public statement; the account of its response is the broadcaster's. Why it matters: a report is a request judged against a policy, and if the reviewer reads it differently the answer is no. When that happens, filing the same report again tends to get the same answer, so the next move is a different basis, not a repeat.
Source: read the report.
12 September 2026: Creators find their faces in ads for products they never endorsed
A newspaper reported that several creators found AI fakes of themselves in ads for weight loss drugs, cosmetics and blood testing they never agreed to. The social media company running the ads took weeks to remove one creator's ad, and a makeup ad stayed in circulation after multiple removal requests. A second outlet, building on that reporting on 16 September, described fakes built from creators' old selfies and vlogs. Why it matters: a public archive is enough to build a fake, and a brand partner cannot easily tell a fake endorsement from a real one.
Sources: the original report and the follow up report.
Related topics
Impersonation and AI likeness rights: fake accounts, cloned voices and the rights that protect your name and face.
Leaked and stolen content: the legal routes for getting leaked paid content taken down.
Copyright enforcement and site blocking: how courts and regulators are blocking sites that ignore takedown notices.
Laws, court rulings and platform decisions are moving, but the copies do not wait for any of them. If deepfakes of you or someone you manage are online now, see how Ceartas deepfake removal works.
