The USCO Can Register AI Content. It Can't Stop Deepfakes. What Congress Must Do Next.

Two things happened in the same congressional hearing room on May 12, 2026.

The Register of Copyrights confirmed the U.S. Copyright Office has built a functional system for registering AI-assisted creative work. And she told Congress that system can’t protect anyone from what deepfake technology does. Those are two different problems requiring two different solutions, and so far, Congress has delivered neither.

The May 12 Hearing: What Perlmutter Said and What It Signals

Shira Perlmutter appeared before the Senate Committee on the Judiciary Subcommittee on Intellectual Property for an Oversight Hearing. The subcommittee’s oversight mandate covers copyright law and the USCO’s operations. The hearing covered two substantive areas: AI and copyright, and *Cox v. Sony*.

According to Holland & Knight’s analysis of the testimony, Perlmutter disclosed that the Office has registered over 7,000 claims containing disclaimed AI-generated elements. The disclosure method matters: creators identify the AI-generated portions of their work, disclaim them, and receive copyright protection for the human-authored portions. The practice is established. The volume, 7,000+ claims, means this isn’t an edge case. The USCO is processing AI-element registrations routinely, at scale. That’s a meaningful operational development for any compliance team building an AI content workflow.

But the 7,000+ figure comes with a ceiling. Registration protects human-authored expression. It doesn’t protect against someone using your likeness in a deepfake. It doesn’t address the voice replication scenario. It doesn’t resolve what happens when a digital replica of a real person is created without consent and distributed. Those are digital replica problems, and current federal copyright law doesn’t reach them adequately. According to Holland & Knight’s analysis, Perlmutter told the Senate that federal legislation addressing digital replicas is urgently needed, consistent with the framework established in the Copyright Office’s Copyright and Artificial Intelligence report.

The report’s conclusions on digital replicas are publicly established. The legislation hasn’t been.

The Cox v. Sony Gap: What the Supreme Court Left Behind

The second major topic at the May 12 hearing was *Cox v. Sony*, a Supreme Court decision that Holland & Knight’s May 22 analysis frames as leaving a legislative gap Congress must address. The H&K article’s title makes the problem explicit: “Can Congress Fill the Cox v. Sony Sized Hole in Copyright Law?”

The specific contours of what *Cox v. Sony* decided are described in Holland & Knight’s analysis of Perlmutter’s testimony. What the hearing record makes clear, from H&K’s coverage, is that the USCO views the decision as having created legislative work that Congress now needs to do. Perlmutter identified the gap. She didn’t close it.

Secondary liability in copyright law determines when a company that enables infringement can itself be held liable. For AI platforms and tool developers, secondary liability is the exposure question that keeps general counsel awake. *Cox v. Sony* adjusted the contours of that question. Congress, per the USCO’s testimony, now needs to decide what the new contours should be.

Connecting the Threads: Sahni, SCOTUS, and Perlmutter as a Single Pressure Point

The Sahni lawsuit, the SCOTUS cert denial, and the Perlmutter testimony are three manifestations of the same structural problem: copyright law was built before generative AI existed, and the system is now being asked to handle questions it wasn’t designed to answer.

The SCOTUS cert denial (May 16, 2026) left the lower court’s human authorship ruling intact. The Court didn’t create new law, it preserved existing precedent. The Sahni lawsuit is testing whether that existing precedent applies correctly to AI-assisted work where human creative decisions drove the process. If the Central District of California rules in Sahni’s favor, the USCO’s current standard gets challenged. If it rules against him, the standard is validated, but only until the next case tests a different edge.

Perlmutter’s testimony sits above all of this. The Register of Copyrights isn’t just reporting on cases and registrations. She’s mapping the gaps in the existing framework and asking Congress to fill them. The 7,000+ registration figure is evidence that the USCO is managing the present. The deepfake legislation request is a signal that the present is already insufficient for the near future.

The Compliance Environment: What Two Different Problems Require

Don’t expect a single legislative fix. The USCO is identifying at least three distinct gaps: AI-element authorship (partially addressed through the disclaim-and-register practice), digital replica protection (not addressed by current law), and secondary liability contours post-*Cox v. Sony* (addressed by the Supreme Court in a way that may require congressional correction). Those gaps require different legislative vehicles.

For compliance teams at companies creating AI-assisted content, the near-term posture is clear: build your registration practice around the disclaim-and-separate model the USCO has operationalized. Document human creative decisions at each step of the AI workflow. Register promptly. The Office is processing these claims; don’t leave your human-authored expression unprotected.

For companies operating in spaces where digital replica risk is relevant, entertainment, talent platforms, voice AI, synthetic media, the near-term posture is harder. Federal legislation doesn’t exist yet. The FTC’s Take It Down Act enforcement (active since May 2026) addresses a narrow slice of the problem. State law, California’s in particular – provides some protection. But the federal framework Perlmutter described as urgently needed hasn’t been built.

What to Watch

The legislative calendar is the primary signal. The Subcommittee on Intellectual Property held an oversight hearing. The next step toward legislation is a markup, where the subcommittee drafts and votes on a bill. Watch for any markup announcement as the trigger that moves digital replica protection from a policy proposal to an actual legislative vehicle.

The *Cox v. Sony* gap is on a parallel track. If Congress treats it as a technical copyright fix, rather than a broader AI liability question, it could move faster than digital replica legislation. Watch for any bill text that specifically addresses secondary liability in the context of AI-generated content.

The Sahni case will produce a federal court ruling that articulates the human authorship standard more precisely than the USCO has in its administrative practice. If that ruling arrives before Congress acts, it becomes the operative threshold. If Congress acts first on the authorship question, the statute supersedes the case.

TJS Synthesis

The USCO has done something real: 7,000+ registrations at scale, a functioning disclosure framework, a clear articulation of where the law falls short. The agency is ahead of Congress. Perlmutter gave the Senate a clear map on May 12, authorship is manageable, digital replicas are not, *Cox v. Sony* needs a legislative fix. What happens next depends entirely on whether Congress treats those as three separate items requiring three separate processes, or recognizes that they’re one underlying problem requiring a coordinated response. The history of copyright legislation suggests the former. Compliance teams should plan accordingly: build for the world where registration works, digital replica protection is still absent, and secondary liability remains unsettled, because that’s the world that exists right now, and Congress won’t resolve all three gaps before the next AI copyright case reaches a federal courtroom.