What Does the TAKE IT DOWN Act Require for Deepfake Removal?

What does the TAKE IT DOWN Act require?

The Tools to Address Known Exploitation by Immobilizing Technological Deepfakes on Websites and Networks Act, known as the TAKE IT DOWN Act, became Public Law 119-12 on 19 May 2025. It does two things. Section 2 amends section 223 of the Communications Act of 1934 to create federal crimes for knowingly publishing nonconsensual intimate visual depictions through an interactive computer service, covering both authentic images and digital forgeries. Section 3 creates a notice-and-removal duty for covered platforms, enforced by the Federal Trade Commission.

The criminal provisions took effect on enactment. The platform provisions gave covered platforms one year, to 19 May 2026, to establish the process. Both apply to AI-generated content in the same way as to real photos, which is why the Act is usually described as the first US federal law aimed squarely at sexual deepfakes.

What counts as a digital forgery under the Act?

The statute defines a digital forgery as any intimate visual depiction of an identifiable individual created through software, machine learning, artificial intelligence or other computer-generated or technological means, including by adapting or altering an authentic depiction, that, viewed as a whole by a reasonable person, is indistinguishable from an authentic visual depiction of that individual. Face-swapped explicit videos, AI "undressing" images and fully synthetic nudes of a real, identifiable person can all meet that test.

For adult victims, publishing a digital forgery is an offence when it was published without consent, depicts something the person did not voluntarily expose in a public or commercial setting, is not a matter of public concern, and either was intended to cause harm or caused harm, including psychological, financial or reputational harm. The Act states that consenting to the creation of an image is not consent to its publication. Offences involving minors have a separate standard. The statute carves out disclosures to law enforcement, in legal proceedings, for medical or scientific purposes and for reporting unlawful content.

How does the 48-hour notice-and-removal process work?

A covered platform is a public-facing website, online service or app that primarily provides a forum for user-generated content such as messages, videos, images, games and audio, or that regularly publishes or hosts nonconsensual intimate content as part of its business. Broadband providers and email are excluded, as are services whose content is mainly preselected by the provider with only incidental interactive features.

Each covered platform must post a clear and conspicuous, plain-language notice explaining how to submit a request. A valid request comes from the depicted person or someone authorised to act for them and contains four elements: a physical or electronic signature, information reasonably sufficient to identify and locate the depiction, a brief statement of good-faith belief that it is not consensual, and contact information. Once it receives a valid request, the platform must remove the depiction as soon as possible and no later than 48 hours, and make reasonable efforts to identify and remove known identical copies.

Failure to reasonably comply is treated as a violation of an FTC rule on unfair or deceptive acts, with the Commission's usual penalties, and the Act extends FTC jurisdiction for this purpose to entities it would not normally reach, such as non-profits. The Act also shields platforms from liability for good-faith removal of material claimed to be nonconsensual, even if it later proves lawful.

• Publish the request process and staff intake so the 48-hour clock can be met at any hour.
• Validate the four required elements and log the time a valid request was received.
• Remove the content, then search for known identical copies with hash matching and record what was found.
• Keep evidence secure and access-limited, because the material is intimate and may be needed by law enforcement.
• Track disputed or abusive requests separately, since the good-faith liability shield favours removal.

Where does deepfake detection help under the TAKE IT DOWN Act, and where does it not?

The legal trigger is a valid request, not a finding that an image is fake. Authentic intimate images are covered too, so a platform should never delay removal while it tries to prove synthesis. Detection cannot establish identity, consent, age, publication intent or whether the content is a matter of public concern, and none of those can wait for a model score.

Detection earns its place around the edges of the process. It helps a trust and safety team triage incoming reports, spot related synthetic content the requester did not list, and recognise re-uploads that are not byte-identical, since edited or re-encoded copies defeat exact hashes. It also helps with records: a forensic analysis attached to the case explains why a team treated an item as a forgery, which matters for law enforcement referrals and for victims pursuing civil remedies. And it supports proactive moderation where a platform's own policy bans sexual deepfakes regardless of requests.

Victim-side tools sit alongside. Services such as StopNCII.org let people create a hash, a digital fingerprint, of an intimate image on their own device; only the hash is shared, so participating platforms can block matching uploads without ever receiving the image. Our guide on how to report synthetic intimate imagery covers the steps a victim can take.

How can NeuroDetect support TAKE IT DOWN Act workflows?

NeuroDetect, Neural Defend's forensic media workspace, fits the parts of the process where a person needs evidence they can explain. A reviewer uploads an authorized image or video into a case. The analysis keeps model and version context, and explainability heatmaps and attention views show the regions or frames that indicate face manipulation or generative synthesis. An LLM-assisted summary describes the findings in plain language linked to the evidence, and the reviewer makes the final call. The downloadable forensic report records observations, uncertainty, provenance, model context and analyst notes, which is the kind of record a law enforcement referral or a civil claim needs.

For volume screening, such as checking re-uploads or proactively scanning images against a platform policy, NeuroVerify returns a unified risk score with signal-level evidence from six signals, including Face Manipulation Detection, AI Generative Model Analysis and Visual Safety Screening, in under 200 ms at p95 for eligible image requests under defined test conditions. On the public VendorBench-100 benchmark of 100 adversarial still images, including face swaps and AI photo edits, Neural Defend ranked first of 36 systems with an MCC of 0.876. Removal decisions under the Act should still follow the request process, not the score.

Because the content is intimate, deployment matters. Neural Defend can run as a managed cloud API, in a private cloud or VPC, as a secure container, on-premises or air-gapped, and holds ISO/IEC 27001, ISO/IEC 27701 and SOC 2 Type II certifications.