The Copyright Office registers AI-assisted work. It just can’t protect you from what someone else’s AI does with your face.
That’s the gap Shira Perlmutter put on the Senate record on May 12, 2026. Perlmutter, Register of Copyrights, appeared before the Senate Committee on the Judiciary Subcommittee on Intellectual Property for an Oversight Hearing. According to Holland & Knight’s analysis of the hearing, Perlmutter disclosed that the Office has registered over 7,000 claims containing disclaimed AI-generated elements, filings where creators identified which portions were AI-generated and which were human-authored, with copyright attaching to the human portions. The Office is processing those claims. The system is working, at least for the authorship question.
The deepfake question is different.
According to Holland & Knight’s analysis, Perlmutter reiterated that federal legislation addressing digital replicas is urgently needed, consistent with the conclusions of the Office’s Copyright and Artificial Intelligence report. The report’s framework on digital replicas is publicly established; the legislation to implement it hasn’t materialized. Congress hasn’t passed a bill. No specific bill text exists for the digital replica protection Perlmutter is asking for.
US AI Copyright Legislative Landscape, May 2026
The hearing also addressed *Cox v. Sony*, the recently decided Supreme Court decision that Holland & Knight’s May 22 analysis identifies as creating a legislative gap Congress must address. The H&K article’s title frames it directly: “Can Congress Fill the Cox v. Sony Sized Hole in Copyright Law?” The USCO is telling Congress the answer is yes, and that the cost of not doing so is measurable.
Why this matters for compliance teams
The 7,000+ figure isn’t just a data point. It tells you the USCO has operationalized AI-element registration at scale. Compliance teams at companies creating AI-assisted content should understand what that means: the Office will register your work if you disclose and disclaim the AI portions correctly. That’s registrable. What’s not protected is the deepfake scenario, someone else using your likeness or voice without consent. That protection doesn’t exist in federal law yet.
The catch is that most companies are building AI content workflows assuming the legal framework will catch up. Perlmutter’s testimony suggests the framework is catching up on authorship, and falling further behind on digital replicas. Those are two different compliance problems.
What to Watch
What to watch
The legislative calendar matters here. The FTC’s Take It Down Act enforcement (active since May 2026) addresses a narrow slice of the digital replica problem. Federal legislation on digital replicas would need to go further. Watch for whether the Subcommittee on Intellectual Property moves from oversight to markup, a markup would signal Congress is ready to legislate, not just hear testimony. The Cox v. Sony gap also creates pressure for a legislative fix on the secondary liability side. Perlmutter gave Congress a clear roadmap. Whether Congress follows it is the open question.
The USCO is telling Congress it has the registration infrastructure. It doesn’t have the statutory authority to fix what deepfakes break. That authority requires legislation – and legislation requires votes.