Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Shira Perlmutter received an email terminating her as U.S. Register of Copyrights on May 10, 2025, one day after the Copyright Office released a report examining whether generative-AI companies may use copyrighted works to train models without permission. The timing prompted allegations that she was punished for the report. It establishes a close sequence, not proof of motive: the administration disputed that explanation, and a separate legal fight emerged over who had authority to remove her.
Who was Shira Perlmutter?
Perlmutter became Register of Copyrights in 2020, appointed by Librarian of Congress Carla Hayden. The Register leads the U.S. Copyright Office, which administers copyright functions and advises Congress on copyright policy. The Office sits within the Library of Congress, a legislative-branch institution—a fact that became central to the dispute over Perlmutter’s removal.
The sequence was unusually compressed. President Donald Trump removed Hayden on May 8, 2025. The Copyright Office released a pre-publication version of its AI-training report on May 9. On May 10, the White House Presidential Personnel Office notified Perlmutter by email that her termination was effective immediately. Paul Perkins was designated acting Register after the removal attempt. Perlmutter and lawmakers argued that the timing suggested retaliation; the administration offered other explanations, including allegations about her conduct at the Library of Congress. Those explanations are competing claims, not established findings.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the AI-training report actually said
The report asked whether copying copyrighted works to develop and train generative-AI systems may infringe copyright, and when the fair-use defense might apply. Its conclusion was conditional, not a blanket verdict: training can involve acts that implicate copyright owners’ exclusive rights, but fair use depends on the facts of a particular use. The report did not declare all AI training illegal, prohibit AI development, or require every company to license every item of training data.
#1 Best Overall
Fair use is a legal doctrine assessed under four statutory factors, including the purpose and character of the use, the nature of the work, how much was used, and the effect on the work’s market. The Copyright Office’s fair-use overview explains the doctrine; courts decide individual disputes. The Office’s report was policy analysis, not a court judgment, statute, or binding regulation.
The Office acknowledged the public benefits of AI and the possibility that training involves transformative technical processes. It also said it could not conclude that unlicensed training automatically receives a special copyright-related fair-use benefit beyond the ordinary analysis. The source and legality of the data, the type of works copied, the purpose of the use, the model and its outputs, and market effects can all matter. Commercial use is relevant, but does not alone settle the question.
Why market effects and licensing mattered
The report focused on potential harm when generated outputs substitute for works in the training data or compete with markets for them. A model that memorizes and reproduces protected material raises a different concern from one whose outputs are not substantially similar to any particular work. The Office also discussed the possibility that AI-generated material could dilute a market for creative work, even where a specific output does not copy a particular work. Such market concerns do not automatically establish infringement; they are part of a broader, fact-specific analysis.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsRank #2
The report noted that the existence or likely development of licensing markets can weigh against fair use under the market-effect factor, and that voluntary licensing was already occurring in some sectors. It also recognized the practical difficulty of licensing every work that might be used in training. Courts, Congress, and the market had not resolved how those competing considerations should apply across different datasets and systems.
In this context, headlines about “AI abuse” are shorthand, not an official legal category. The report’s concerns included mass copying, uncompensated use, possible substitution and market dilution, and uncertainty about licensing. Its countervailing considerations included AI’s potential benefits, the technical and economic difficulty of comprehensive licensing, and the fact that outcomes may differ according to data access, training methods, and model outputs. It neither declared AI inherently abusive nor endorsed unrestricted copying.
The report was part of a larger AI policy project
The training report was the third part of the Copyright Office’s AI initiative. Part 1, released July 31, 2024, addressed digital replicas; Part 2, released January 29, 2025, examined copyrightability of AI-generated outputs; and Part 3, dated May 9, 2025, addressed training. The Office’s general position on authorship is that purely AI-generated material is not copyrightable, while human-authored work made with AI assistance may qualify when it contains sufficient human creative contribution. In testimony for a May 2026 Senate hearing, Perlmutter said the Office had registered more than 7,000 claims containing AI-generated material where that material was appropriately disclaimed. That figure does not mean purely AI-generated works received copyright protection.
The May 9 document was expressly a pre-publication version. The Office said it expected to publish a final version later and did not expect substantive changes to its analysis or conclusions. That status does not make the report irrelevant, but it is another reason not to treat it as a binding legal decision. The Office’s AI initiative page describes the reports and their status.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Why the removal became a legal dispute
The controversy involved two questions that should not be collapsed into one: whether the report prompted the firing, and whether the President could lawfully remove the Register directly. Perlmutter argued that the Register is a legislative-branch official appointed and supervised by the Librarian of Congress, and that the President lacked statutory authority to remove her. The administration argued that the President had authority to remove or direct her removal and that the attempted termination was valid.
A district court denied Perlmutter’s request for a preliminary injunction on May 28, 2025. On September 10, a divided D.C. Circuit panel temporarily blocked the removal. Its reasoning emphasized the statutory structure: the Register is appointed and supervised by the Librarian, and the panel concluded that the Librarian—not the President—had removal authority. That was preliminary relief while the case proceeded, not a final resolution of every legal issue or a finding that the firing was retaliation.
Rank #4
The legal question about who may remove a Register is distinct from the merits of any AI copyright dispute. Even if a court ultimately resolves the institutional question, that would not determine whether a particular company’s training practices infringe copyright or qualify as fair use.
Where the matter stood in 2026
Senate Judiciary materials for a May 12, 2026 oversight hearing listed Perlmutter as Register of Copyrights and director, and the Copyright Office published testimony under her name. That is evidence that she was continuing to function in the role or was being treated as the lawful incumbent at that point. It does not, by itself, settle the case or establish its ultimate merits outcome.
Recommended Free Tools
What the episode means for creators and AI developers
For creators, the report is not a guarantee that a work cannot be used in training, nor does it create a universal right to payment. Copyright registration, licensing, platform terms, and litigation are separate matters. For a commercially valuable work, keeping dated originals, drafts, source files, and records of human creative decisions can help document authorship and development. Creators should review the terms of services they use and avoid assuming that an opt-out setting or a robots.txt instruction alone provides a complete legal shield. Disputes and licensing decisions may call for advice from a copyright lawyer.
For AI developers, the report is not a safe harbor or an automatic ban. Its analysis underscores why dataset provenance and access methods matter, and why it can be useful to distinguish licensed, public-domain, user-submitted, and disputed material. Companies may also assess memorization and output-substitution risks and consider whether licensing approaches fit particular sectors. These are practical risk-management considerations, not legal conclusions that the report imposes on every developer.
For both groups, the unresolved questions reach beyond the individual case: how copyright law should balance creative markets with AI’s potential benefits, what licensing arrangements are feasible, and how much independence the office advising Congress should have from the executive branch. The report put those questions into sharp relief; it did not answer them all.
Quick Recap
Key dates
| Date | Event |
|---|---|
| October 2020 | Perlmutter becomes Register of Copyrights. |
| July 31, 2024 | The Copyright Office releases Part 1 of its AI reports, on digital replicas. |
| January 29, 2025 | Part 2 addresses copyrightability of AI-generated outputs. |
| May 8, 2025 | Trump removes Librarian of Congress Carla Hayden. |
| May 9, 2025 | The Office releases the pre-publication AI-training report. |
| May 10, 2025 | Perlmutter receives an email terminating her immediately. |
| May 28, 2025 | A district court denies her request for a preliminary injunction. |
| September 10, 2025 | A divided D.C. Circuit panel temporarily blocks the removal. |
| May 12, 2026 | Senate Judiciary materials list Perlmutter as Register for an oversight hearing and publish her testimony. |
Sources and further reading
- Copyright Office, Part 3 report on generative-AI training (pre-publication version)
- D.C. Circuit order concerning Perlmutter’s removal
- Supreme Court filing containing the district-court record
- Senate Judiciary oversight hearing, May 12, 2026
- Perlmutter’s testimony for the hearing
- Senators’ statement criticizing the removal
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →
