The Copyright Office told applicants to disclose AI content

The picture

On 16 March 2023 the U.S. Copyright Office published formal registration guidance on works containing AI-generated material in the Federal Register, 88 Fed. Reg. 16,190. The notice responds to what the Office describes as a rise in applications that either did not disclose AI involvement or mentioned an AI tool only in a title or acknowledgment. It states that 'applicants have a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author's contributions to the work,' and it directs holders of existing registrations to check whether their filings already meet that standard.

What the documents show

The Federal Register notice is the primary legal document: dated, numbered, and published under the Office's own authority, it is the guidance itself rather than a description of it. The Office's own AI resource page confirms the same document and date and cross-references it against the Office's later registration decisions, including the Zarya of the Dawn cancellation eleven months afterward, showing the guidance was applied rather than left as an unenforced statement. Read together, the two sources establish that the disclosure duty predates and anticipates the specific cases that later tested it, rather than being written in response to any one of them.

What it is allowed to decide

This is Copyright Office policy, not adjudication; it cannot itself decide a class or authority for any artifact, but it sets a documentation requirement previs work now sits inside. It establishes Provenance as the authority a registration applicant must affirmatively assert, since the guidance requires naming what a human did and what a machine did. It says nothing about Dimensional, Optical, Photometric or Physical authority, and it does not itself determine how much human editing is enough, a line the guidance leaves to case-by-case examination.

The disclosure label

Any previs asset carrying AI-generated elements toward a registered deliverable would need a label reading: AI-generated content present and must be disclosed under Copyright Office policy dated 16 March 2023; the class of the underlying registration is limited to the human-authored elements unless the human contribution is separately described; Provenance held only for the disclosed human portions; dated 16 September 2026 and properly asserted by the production's copyright counsel at filing, not left to the previs vendor.

• Does the production's registration application name which elements were AI-generated and which were human-made?
• Has anyone checked an existing registration against this guidance since it was filed?
• Is the human contribution described in enough detail that an examiner could evaluate it, rather than merely asserted?

The guidance is three years old as of this entry and has already been tested against a real registration; a previs desk that ignores it is relying on an untested assumption rather than a documented one.