Journalism has never had a legal definition of what an editor is. The role has been defined by practice, by newsroom tradition, by professional norms that developed across more than a century of institutional press culture — and by the understood but unwritten premise that the person whose name appears on a piece of writing has exercised some form of judgment over it. That premise was enough for as long as editors were humans by default. The NY FAIR News Act, passed by the New York Legislature in 2026 and awaiting Governor Hochul’s signature, is the first time a government has tried to codify what that premise actually requires — to write into statute a definition of what human editorial presence means, and what it is legally sufficient to claim.
The law’s core requirement is straightforward: The law’s core requirement has two parts, not one: any news content ‘substantially composed, authored, or created’ through generative AI must carry a disclosure to that effect, and, separately, any content created in whole or material part by generative AI must be reviewed by a human employee with editorial authority before publication. The two obligations apply together — disclosure does not excuse a publisher from the review requirement, and review does not excuse the disclosure requirement. The disclosure-or-review structure is the mechanism by which the law creates accountability. It tells publishers: you may use AI to generate news content, but you must either tell readers you have done so, or you must put a human in the loop who is responsible for the output. What it does not tell publishers — and what the law’s drafters presumably had to leave unresolved — is exactly what being in that loop requires.
What “editorial control” means when it has to mean something specific
The phrase the law uses is “human employee with direct editorial control.” In a working newsroom, this phrase would be understood intuitively: the editor who assigns the story, reads the draft, pushes back on sourcing, changes the lede, and approves the final version is exercising direct editorial control. No one would dispute it. The problem is that the intuitive understanding was built for a workflow in which humans wrote the drafts being edited. It does not specify, with the precision a statute requires, what the reviewing human must actually do to qualify.
This ambiguity is not a drafting failure. It is a genuine conceptual problem that the legislation surfaces rather than creates. What minimum engagement constitutes review? Does reading the output once satisfy the requirement? Does the reviewer need to change something, or is approval without modification sufficient? Can a single editor review fifty AI-generated stories in a shift and be said to have exercised direct editorial control over each of them? These questions do not have obvious answers, and they matter because the answer determines whether the law functions as a meaningful accountability mechanism or as a procedural checkbox that AI-dependent publishers satisfy by routing output through a nominal human review that changes nothing.
Sponsors Senator Patricia Fahy and Assemblymember Nily Rozic, both Democrats, have framed the law primarily as a labor and transparency measure — a protection for journalists whose jobs are at risk and a guarantee for readers who deserve to know what generated the content they’re reading. Those goals are clear and defensible. But the law’s operational effect depends on what “direct editorial control” turns out to mean in enforcement, and that question will almost certainly be answered case by case rather than in the statute’s text.
The labor dimension the law actually addresses
The disclosure requirement gets most of the attention in coverage of the NY FAIR News Act, but the labor provisions are, in some respects, the more structurally significant part. The law restricts news organizations from firing journalists or reducing their pay and benefits as a result of AI adoption. It also includes protections for confidential source material — provisions designed to prevent AI systems from ingesting the protected information that journalists gather in confidence, which could expose sources through downstream outputs in ways that are difficult to trace and harder to defend.
The NewsGuild of New York, which supported the bill, has been explicit about this dimension: the law is in part a floor against the specific displacement risk that AI-generated content represents for the journalists who currently produce it. A publisher who can generate news with AI at lower cost than employing reporters has a structural incentive to reduce its reporter headcount. The law’s labor protections are an attempt to interrupt that incentive while the industry, regulators, and the public work out what AI-generated news actually means for journalism as a practice.
Whether those protections will hold under pressure is a different question. Labor protections that prohibit AI-driven layoffs are genuinely novel, and their enforceability depends on whether “a direct result of AI adoption” can be established in cases where publishers have access to multiple plausible explanations for staffing decisions. The WGA East, which also supported the legislation, urged Governor Hochul to sign it quickly, calling it a way to “place value on the vital work done every day by newsroom workers.” Whether other states follow New York’s approach is, at this point, the article’s own speculation rather than a claim any coalition member has made.
How robustly the protections function in New York will determine whether that template is worth replicating.
What the law doesn’t resolve about AI and authorship
The deeper question the NY FAIR News Act engages without fully answering is what authorship means when generative AI is involved in producing a text. The law’s disclosure trigger — content “substantially composed, authored, or created” by AI — requires a determination about how much of a piece of writing AI generated before the disclosure obligation attaches. This threshold question is one that publishing, copyright law, and academic integrity policy have all been wrestling with since large language models became capable of producing plausible prose, and none of them has resolved it.
In practice, the AI-in-news workflow is rarely pure generation — a journalist prompts an AI system, edits its output, supplements it with original reporting, and produces something that is neither entirely AI-generated nor entirely human-written. Whether the resulting piece triggers the disclosure requirement, requires a human reviewer with editorial control, or falls outside the law’s scope entirely depends on which parts of the process are attributed to AI and how “substantially” is eventually interpreted. These are not edge cases. They describe the predominant workflow at the news organizations most likely to be using AI at scale.
The copyright questions adjacent to this are unresolved at the federal level in ways that complicate state-level efforts to define the human contribution. The US Copyright Office has taken the position that AI-generated content without meaningful human creative control is not copyrightable, but has declined to specify what level of human intervention is sufficient to cross that threshold. New York’s law is making a related determination in a different register — not about copyright eligibility but about editorial accountability — and it faces the same definitional difficulty: specifying the human contribution precisely enough to be enforceable without being so prescriptive that it doesn’t match how newsrooms actually function.
Why the ambiguity may be the point
One interpretation of the law’s open questions is that they represent weaknesses — places where publishers will find room to comply in form while evading the spirit of the requirements. Another interpretation is that the ambiguity is deliberate and appropriate: that establishing the principle of human editorial responsibility for AI-generated news content is the primary achievement, and that the specifics will be worked out through enforcement, litigation, and the development of industry norms that the law creates incentives to form.
Disclosure requirements in media have historically operated this way. The FTC’s endorsement disclosure rules, for example, established the principle that undisclosed paid promotions are deceptive without specifying exactly what adequate disclosure looks like in every format and context. That specification has been developed over years of guidance, enforcement actions, and industry practice. The NY FAIR News Act’s definition of editorial control may follow a similar path — the statute establishes that a human reviewer with direct editorial control is required; enforcement actions and litigation will eventually establish what direct editorial control actually requires, and the industry will adapt accordingly.
What is not ambiguous is the precedent. Before this law, no government had attempted to define, as a matter of enforceable public policy, what a human editor’s presence in a piece of writing requires. The question had been left to newsrooms, to journalism ethics codes, to professional norms. New York has decided that those mechanisms are insufficient for the current moment — that the speed and scale at which AI can generate news content, and the commercial incentives that make AI-generated news attractive to resource-constrained publishers, require a legal floor rather than a professional one.
Whether that floor is set at the right height, and whether it can be enforced in a way that makes it meaningful, are open questions that will be answered after Governor Hochul signs or declines to sign the bill. What the law has already done — regardless of what happens next — is establish that the question of what a human editor does is no longer purely a professional one. It is now, at least in New York, a legal one. That shift — from professional norm to legal standard — is what other newsrooms, other legislatures, and publishers building AI-assisted workflows will need to reckon with.
