S. 5057: Safeguarding Against Fabricated Exploitation Through Artificial Intelligence Act of 2026

This bill would restrict federal agencies from buying or using certain artificial intelligence models that can generate realistic images or video, if those models are found to be associated with harmful synthetic sexual content.

What counts as a prohibited AI model

The bill defines a covered application as an AI model that is determined to be out of compliance with standards meant to reduce:

• non-consensual intimate imagery,
• synthetic child sexual abuse material, or
• child pornography or non-consensual intimate depictions of adults.

An AI model can also be treated as covered if a federal court determines that it has generated child pornography or non-consensual intimate depictions of an identifiable adult.

New standards and testing

The National Institute of Standards and Technology (NIST) would be directed to start, within 90 days, creating performance benchmarks aimed at preventing AI from generating obscene, degrading, or abusive content, including synthetic child sexual abuse material and non-consensual intimate images of adults. NIST would also have to update those benchmarks over time as technology and evasion methods change.

Within 180 days, NIST would also create a voluntary testing program for vendors, based on those benchmarks.

Limits on federal use

Federal agencies would be prohibited from purchasing or using a covered application on government or contractor information systems.

Agencies would generally have to remove covered applications from their systems within 180 days after enactment. If a model is later determined to be covered, agencies would have 180 days after that determination to remove it as well.

Ongoing review of products

Starting one year after enactment, the General Services Administration would have to review products and services in its purchasing programs at least every 90 days to see whether they include a covered application. Federal agencies would also have to review their own products and services at least once a year to check for covered applications.

Possible exceptions and safeguards

If a vendor cannot modify a model to remove the problem, an agency could instead apply additional technical and compliance safeguards to prevent misuse. If it does this, the agency would have to publicly certify within 30 days that the safeguards are enough to prevent the model from being used to generate child pornography or non-consensual intimate depictions of an identifiable adult, and notify Congress within 7 days after that certification.

Agencies could also give vendors a chance to fix the model at the vendor’s expense so it no longer counts as a covered application.

Restrictions on circumvention tools

The bill would make it unlawful to deliberately make, import, or sell tools or services mainly designed to bypass, remove, or tamper with the technical controls used to stop an AI model from generating child pornography. It also covers products marketed for that purpose or with little other practical use.

Private lawsuits

The bill would create a private right of action, meaning certain people could sue in federal court. This includes:

• an individual whose image or likeness appears in child pornography,
• a person harmed by a developer’s failure to use strong enough controls to stop their image from being used this way,
• and a developer whose controls were bypassed in violation of the bill.

Courts could issue injunctions, impound devices or products involved in violations, award actual damages or $50,000 in statutory damages, and in some cases increase damages for repeat violations. The bill says courts may not impose a prior restraint on speech or the press protected by the First Amendment.

Relevant Companies

• None found