S 1396
Content Origin Protection and Integrity from Edited and Deepfaked Media Act of 2025
TL;DR
Senator Maria Cantwell's COPIED Act would require AI companies to embed tamper-resistant content provenance markers (like watermarks) in AI-generated images, video, audio, and text, and would ban removing those markers or using journalistic and artistic content to train AI without permission. It gives creators, the FTC, and state attorneys general the power to sue violators.
How This Might Impact Your Business
Companies building generative AI tools (image, video, audio, text) would have to attach machine-readable provenance and watermarking information to outputs within 2 years of NIST setting standards.
Platforms distributing synthetic content (social media, stock media sites, streaming services) would need systems to detect and preserve provenance data rather than stripping it out.
Using copyrighted news articles, music, film, photography, or other creative works to train AI models without the owner's consent would become illegal, opening the door to lawsuits from publishers, studios, labels, and individual creators.
Enforcement comes from the FTC (treating violations as unfair or deceptive practices), state attorneys general, and a private right of action for content owners, meaning statutory damages and injunctions are on the table.
Small research projects and personal, non-commercial use appear to get lighter treatment, but any commercial AI product is squarely in scope regardless of company size.
NIST would set the technical standards, so companies should expect a compliance clock starting once those standards publish, not when the bill passes.
Sectors relying on scraped web data for model training (foundation model developers, ad tech, marketing AI) face the biggest business model disruption.
What Should You Do
Inventory every AI model your company trains or fine-tunes and document the data sources; flag any use of news, music, film, images, or other copyrighted material without a license.
Have your product and engineering teams evaluate C2PA and similar content provenance standards now so you are not scrambling when NIST issues requirements.
Ask legal counsel to review your training data licenses and terms of service, and to model the litigation exposure if a private right of action becomes law.
If you distribute user-generated or AI-generated content, assess whether your platform currently strips metadata and start scoping the engineering work to preserve provenance signals.
Track the bill through the Senate Commerce Committee and coordinate with your trade association, as similar provisions may be attached to must-pass legislation.
Who It Affects
Sponsors
Status Timeline
committee
Read twice and referred to the Committee on Commerce, Science, and Transportation.
April 9, 2025