New York’s legislature passed a bill aimed at telling readers when news content has been substantially generated by AI. That is a significant intervention in how publishers present their work, but understanding it requires reading the version that passed. The final bill is a disclosure measure with a copyright-registration exemption. Earlier proposals contained broader newsroom protections that did not survive into the passed text.
What the passed bill would require
Under section 1802 of the passed B version, news content substantially created through generative AI would need a conspicuous disclosure if published, broadcast, disseminated or accessible in New York. Visual content would carry it at the top; audio would carry a spoken disclosure at the beginning. Content eligible for copyright registration would be exempt. The bill would take effect 60 days after becoming law.
That structure makes the origins of content relevant to how it is presented to an audience. It also creates two distinct questions for a publisher assessing a piece: whether AI substantially generated the content, and whether the copyright exemption applies. A workflow label such as “AI-assisted” does not, by itself, answer either question. Publishers would need to look at what happened in the production of the particular work.
The placement requirement matters in practical terms. A disclosure that readers encounter before consuming the content serves a different purpose from an explanation buried on a general policy page. A publisher considering compliance would therefore need to examine the article, image, video or audio experience itself, rather than assume a site-wide statement makes each item’s provenance clear.
What changed from the earlier proposal
The original proposal included compulsory human editorial review, workplace provisions and safeguards for confidential source material. Those provisions are absent from the passed B version. The final bill does not establish a required human-editor role or a combined disclosure-and-review duty, and it does not prohibit AI-driven dismissals or pay reductions.
The gap matters because a bill’s original goals can stay prominent in public discussion long after its operative provisions change. Concern for journalism jobs may explain why the measure was introduced, but the employment protections proposed at introduction are not in the version the legislature approved.
For a newsroom, the distinction also affects planning. Newsrooms should not plan around any required editor workload, such as a set number of AI-generated stories reviewed per shift. Human review remains a valuable editorial practice, but the B version sets no such standard. The legal questions it raises concern the scope of disclosure and the exemption.
Why the copyright exemption deserves attention
Copyright eligibility introduces a question about creative authorship into a measure concerned with transparency. The U.S. Copyright Office’s January 2025 report says AI assistance does not disqualify a work from protection, while purely AI-generated expression is not protected. Human creative contributions to a mixed work may be protectable; their sufficiency is assessed case by case. Prompts alone generally do not establish authorship of the generated output under the technology the Office examined.
In practice, an editor having seen a piece is not the same as the piece being eligible for copyright registration. Reading a draft for accuracy, approving it for publication and contributing original creative expression are different activities. A publisher would need to assess the actual work and human contribution when considering the exemption, rather than rely solely on the job title of the person who signed it off.
Mixed work also requires care. A piece might contain original reporting, human-written passages and material generated by a model. Copyright protection for human expression does not automatically extend to every generated element. How the state exemption would apply to a particular mixed item would turn on its facts; the bill offers no simple percentage-of-editing formula.
What disclosure can and cannot tell readers
A visible label can give a reader information about how content was produced. It cannot establish that the claims in the content are accurate, that every source has been checked or that the item serves its audience well. Those remain editorial responsibilities regardless of whether the production process triggers a legal disclosure.
The same distinction works in the other direction. A piece outside the disclosure requirement would not acquire a guarantee of reliability. Readers still need a clear account of sourcing, named responsibility and an accessible way to raise errors. Publishers should therefore treat provenance as one part of editorial accountability, rather than make a disclosure label carry the entire burden of demonstrating quality.
For the newsroom itself, keeping records of the production process would help answer both internal editorial questions and questions about the exemption. Such records could identify which passages originated in reporting, which were generated, and what the human contribution changed. The bill itself does not require such records, but they would make a workflow far easier to explain.
The NY FAIR News Act’s legislative passage brings public disclosure of substantially AI-generated news into a state policy debate. It leaves questions about employment, confidential sources and the role of human editors unresolved. Those issues still matter deeply to journalists, but the final bill does not address them.
For now, publishers have a concrete text to work from: what the passed version requires, where it stands in the legislative process, and where disclosure ends and copyright eligibility begins. The broader debate about editorial responsibility will be better for starting there.
