Deepfake removal: how to report sexual or defamatory Deepfakes of yourself

01Telling a synthetic image from a real one

In practice you are looking at three kinds of signal, none of them conclusive on its own.

• Internal signals. Hands, teeth, jewellery, hair against a busy background, text in the frame, and the join line at the neck or jaw are where generators and face-swap tools still fail. In video, look at the mouth during fast speech and at the edge of the face when the head turns.
• Provenance signals.Google Search's About this image panel can show when Google first saw a similar image, which pages use it, and, where available, whether the file was captured with a camera or generated with AI. Google states plainly that credit and digital source type metadata can be modified by generation and editing tools, so it may not be accurate.
• Circumstantial signals. The original photograph the forgery was built from is usually still online: a profile picture, a holiday photo, a press shot. Finding the source image and the matching pose is the strongest practical evidence that the sexual version is fabricated.

02Why proving it is fake matters less

The removal grounds do not turn on whether the image is real. They turn on whether you are identifiable in it and whether it was published without your consent.

Google's removal policy for personal sexual content says that whether the content is fake or real, you can request that sexual videos and images involving you are removed from search results. The US offence for synthetic material is written around a digital forgery, which Public Law 119-12 defines as an intimate visual depiction of an identifiable individual created through software, machine learning, artificial intelligence or other computer-generated means, including by adapting or altering an authentic depiction, that when viewed as a whole by a reasonable person is indistinguishable from an authentic depiction of that individual.

In England and Wales the sharing offence in section 66B of the Sexual Offences Act 2003 applies to a photograph or film which shows, or appears to show, another person in an intimate state. That wording is what brings Deepfakes inside the offence without any need to classify the file.

So the question you have to answer in a report is not technical. It is: is this me, and did I agree to it being published? Spend your effort on identifiability and on evidence, not on detection.

03Capture the evidence before you report

Successful reports are removed within hours, which destroys the record. Capture first, report second. Capture everything, do not crop to spare yourself, and never re-upload the material anywhere.

04Report routes for synthetic sexual imagery

Use the route named for fake or manipulated intimate content rather than the general nudity report. The named route is reviewed against a policy that does not require the image to be authentic.

Submit one report per item, not one report listing many. Keep the reference number each route returns; it is what you quote when you escalate or when the item reappears.

05Search engines: cutting the discovery path

Removing a page from a search index does not delete the page, but it is often what actually stops the harm, because almost nobody reaches the host site except through a search for your name.

Google describes two outcomes for a granted request. Full removal means the page no longer appears in Google Search at all. Partial removal means the page no longer appears for any search query containing your name or identifier, but may still appear for other queries. Google also states that for sexual imagery removal requests it does its best to find and remove duplicates from Google Search as standard procedure, and that you can opt out of duplicate removal when you submit the form.

Microsoft uses validated StopNCII.org hashes in Bing Search and, as of May 2026, in its hosted consumer services including GroupMe, Teams Free, OneDrive and Xbox, so a StopNCII.org case feeds the Bing side as well as the platform side.

Search removal and host removal are separate tracks. Run both; neither substitutes for the other.

06Hash blocking, and where it stops

StopNCII.org generates a hash of a file on your own device and sends only the hash. Its guidance states that non-consensual intimate image sharing can include manipulated images and Deepfakes, so a synthetic file is eligible. Participating companies compare uploads against the hash bank and act where the content breaches their intimate image abuse policy.

The limits are structural and you should plan around them.

• It matches files, not faces. The algorithms used are PDQ and PhotoDNA for photographs and MD5 for video. Duplicate copies of the same file share a hash; a newly generated Deepfake of you, or a re-encoded video, does not.
• It only reaches partner platforms. StopNCII.org states it cannot remove images from the whole internet, only the participating platforms listed on its partners page. That list currently includes Microsoft and a set of adult platforms, not every mainstream social network.
• It is forward-looking. If the material is already posted, report it to the platforms first and add the hash for future protection.
• It takes days, not minutes. StopNCII.org says you may need to wait three to five days to see case updates.

Treat hashing as a floor under the problem, not the removal itself.

07United States: the TAKE IT DOWN Act

Public Law 119-12 was enacted on 19 May 2025. It does two separate things.

First, it creates a federal offence. Knowingly publishing a digital forgery of an identifiable adult through an interactive computer service is unlawful where the forgery was published without consent, what is depicted was not voluntarily exposed in a public or commercial setting, it is not a matter of public concern, and publication is intended to cause harm or does cause harm, including psychological, financial or reputational harm. The penalty for an offence involving an adult is a fine, imprisonment for not more than two years, or both; for a minor, not more than three years. Threatening to publish is also an offence.

Second, it imposes a platform duty. A covered platform had to establish a notice and removal process within one year of enactment, that is by 19 May 2026, so the duty is in force now. On receiving a valid removal request from the identifiable individual or an authorised person acting for them, the platform must, as soon as possible and not later than 48 hours, remove the depiction and make reasonable efforts to identify and remove any known identical copies. Failure to comply reasonably is treated as an unfair or deceptive act or practice enforced by the Federal Trade Commission.

A valid request must be in writing and include your signature, physical or electronic; identification of the depiction and information sufficient to locate it; a brief statement of your good faith belief that it is not consensual; and contact details. Write your reports to meet that list even on platforms outside the statute.

08UK: sharing, creating and platform duties

Three distinct points matter, and they commenced on three different dates.

Sharing and threatening to share

Sections 66B to 66D of the Sexual Offences Act 2003, inserted on 31 January 2024 by section 188 of the Online Safety Act 2023, cover intentionally sharing a photograph or film that shows or appears to show another person in an intimate state without consent, and threatening to share one. For a threat, the prosecution does not have to prove the image exists or that it really shows a person in an intimate state. The basic sharing offence is summary only; the variants involving intent to cause alarm, distress or humiliation, or sexual gratification, and the threat offence, carry up to two years on indictment.

Creating and requesting

Sections 66E to 66H were inserted on 6 February 2026 by section 138 of the Data (Use and Access) Act 2025. Section 66E makes it an offence intentionally to create a purported intimate image of another adult without consent, where a purported intimate image is one that appears to be or include a photograph or film of the person but is not, and appears to show them in an intimate state. Section 66F covers intentionally requesting the creation of such an image, or requesting that a particular feature be included or excluded, and applies whether or not the image is ever created and wherever in the world the person asked is located. Both carry summary conviction penalties and both have a reasonable excuse defence.

What platforms must do

Sections 66B, 66E and 66F are listed as priority offences in Schedule 7 to the Online Safety Act 2023: section 66B was added to that list on 8 January 2026 and sections 66E and 66F on 29 June 2026. Priority status means in-scope services carry proactive illegal content duties for this material, so a UK report can be framed as illegal content rather than as a terms of service complaint.

09EU and Australia

Article 50(4) is a labelling duty on whoever deploys the generator. It is not a removal route, but a missing disclosure is a useful fact in a complaint and in any claim. The removal route in the EU is the Article 16 notice, which you should file with the exact URL, a substantiated explanation and a good faith statement.

An Australian complaint can be made by the person depicted, or by someone authorised by them, or by a parent or guardian of a child under 16 or a person who cannot manage their own affairs.

10Who made it, re-uploads, and police

Identifying the poster: the realistic limit

Platforms do not disclose subscriber details to a private complainant. They disclose to law enforcement or under a court order. What you can do yourself is build the record that makes either possible: the handle, the account URL and creation date, posting times, the wording and spelling habits in captions, any other accounts posting the same file within minutes, and the host and registrar of any standalone site. Where the material also appeared on a forum, note the thread and the earliest post, since the first uploader is usually the only one worth pursuing.

Stopping re-uploads

Expect the same file and fresh variants. A monitoring cycle that works has four parts: a fixed set of name, handle and misspelling queries run on a schedule across search engines and the platforms themselves; reverse image search on the source photograph, which tends to surface new forgeries built from it; hash cases kept open so partner platforms keep matching; and a record of every report reference so a reappearance is escalated as a repeat rather than filed as a new complaint. Face-matching monitoring is what catches forgeries that share your likeness but not your file hash, and it is the only layer that scales as new variants are generated.

When to go to police

Go to the police, rather than only to the platforms, where any of these is true: the material involves a person under 18; you are being threatened with publication or asked for money or images; the posting is part of a sustained campaign by an identifiable person; there is any contact with your workplace, school or family; or you are in England or Wales and can point to a section 66B, 66E or 66F offence. Bring the captured record, not the files, and ask for the crime reference number. Keep reporting to platforms in parallel, because a police report does not remove anything.

SARAZIM LLC works the removal and monitoring side of this: evidence capture, the platform and search routes, hash cases, and continuous face-matching monitoring for re-uploads. We are not a law firm and do not act as your legal representative.