US bill wants to make stripping AI watermarks illegal

If you are a photographer, journalist, or digital artist, watching AI models ingest your life's work without permission has felt like an unstoppable train. Congress might finally be laying down some tracks to block it.

A bipartisan trio of senators just introduced the COPIED Act, short for the Content Origin Protection and Integrity from Edited and Deepfaked Media Act. The pitch is straightforward: make AI deepfakes easier to detect, and make stripping digital provenance markers from creative work illegal.

Tamper-proofing the creative supply chain

The bill puts the National Institute of Standards and Technology (NIST) in charge of writing guidelines for content authentication, detection of synthetic media, and watermarking.

It also requires AI tools handling journalistic or creative material to allow users to attach provenance data to their work. Once attached, removing, disabling, or tampering with that information is illegal, outside of narrow exemptions for security research.

Critically, any content carrying this provenance data cannot be used to train AI models without authorization.

Enforcement carries real teeth. The bill hands copyright holders—including newspapers, artists, and broadcasters—the right to sue companies that scrub their markers or use their work without consent. State attorneys general and the Federal Trade Commission would also have enforcement authority.

The legislative effort is led by Senate Commerce Committee Chair Maria Cantwell, along with Senators Martin Heinrich and Marsha Blackburn. While Senate Majority Leader Chuck Schumer set out an overarching AI roadmap, he left specific bills to committees, giving Cantwell's proposal serious institutional weight.

Creative labor groups immediately rallied behind it. Organizations including SAG-AFTRA, the Recording Industry Association of America (RIAA), the News/Media Alliance, and the Artist Rights Alliance released statements backing the measure.

SAG-AFTRA chief negotiator Duncan Crabtree-Ireland noted that accurate digital clones of performers pose an immediate threat to member livelihoods, arguing that the industry urgently needs a transparent and accountable AI supply chain.

Why it matters

For creators and publishers, this bill offers something they have desperately lacked: clear legal standing to hit back against data scrapers. Proving copyright infringement against a multi-modal foundation model in court is agonizingly complex, but proving someone removed your digital tag is far more cut and dry.

For AI companies, it signals that the era of treating the entire public web as free training data is closing fast. If scraping around provenance tags carries federal liability and private lawsuits, developers will have to invest heavily in verifiable data sourcing or face costly litigation.

We think targeting provenance tampering rather than trying to ban specific AI training practices is a clever regulatory lever. It aligns technical standards with legal liability.

The major catch, however, lies in the tech itself. Digital watermarks and metadata are notoriously fragile against simple screenshots, re-encoding, and compression. NIST has a massive engineering challenge ahead, because if the provenance markers cannot survive basic internet distribution, the legal protection could quickly become symbolic.