New York has two AI disclosure bills heading to Governor Hochul’s desk right now. They cover different ground.
A3411B — the Artificial Intelligence Disclosure Act — requires any operator of a generative AI system to display a notice that outputs may be inaccurate. One UI element, 90 days after signing, applies to everyone building AI products.
The NY FAIR News Act (S.8451B / A.8962B) is different in scope, different in mechanism, and considerably more demanding. The Senate passed it 53-7 on June 4, 2026, and the Assembly passed it 130-1 the next day, June 5, 2026, completing passage. It applies specifically to news publishers, but its requirements flow directly upstream to the AI content tools, writing assistants, CMS pipelines, and newsroom automation that builders sell to those publishers.
If you build anything that touches AI-generated news content, this law shapes your product.
What the FAIR News Act Requires
The bill’s four operative mandates are distinct. They cover the output layer, the editorial layer, the internal process layer, and the data layer.
1. Conspicuous AI Authorship Label
Any news content “substantially composed, authored, or otherwise created” through generative AI must “conspicuously imprint on the top of the page, webpage, image, graphic, video or other visual or audio/visual content,” or be verbally stated at the onset of audio, that it “was substantially created by generative artificial intelligence.” Content eligible for copyright registration is exempt. (S.8451B, General Business Law Art. 48 §1802)
This is meaningfully different from A3411B’s generic “outputs may be inaccurate” notice. The FAIR News Act requires affirmative authorship disclosure — stating that a human did not primarily write the piece. For publishers running AI content pipelines, this label must appear at the article level, not just at the product or system level.
Builder implication: AI writing tools that produce article drafts for news publishers need to surface provenance metadata — a flag or field indicating “this output was substantially AI-generated” — so the publisher’s CMS can render the required label. If your tool generates a draft and hands it to a writer who edits one sentence, does that clear “substantially”? The bill leaves this undefined. Your customers will ask. You need an answer in your documentation.
2. Mandatory Human Review Before Publication
No news content “created in whole or in material part by generative artificial intelligence” may be published until it is “reviewed by a human worker who has the authority to approve, deny, or modify any decision recommended or made by the automated system” (bill text, quoted in Nieman Lab; also confirmed in R Street’s opposition commentary). Fully automated content feeds — scheduled news briefs, auto-published sports scores, automated financial summaries — require editorial gating under this law.
Builder implication: If you build content automation tools that publish to news outlets without a human approval step, your product is now non-compliant for New York-based publishers. You need a gating mechanism: a review queue, an approval workflow, a one-click sign-off interface. This is not a cosmetic change — it is a required architectural feature for any tool targeting publishers subject to this law.
The bill does not define what qualifies as “review.” A reasonably defensible standard is that the human must read the content and affirmatively approve it before the publish action executes. Auto-approval after a timer, or an opt-out confirmation screen, would likely not qualify.
3. Newsroom Transparency
Publishers must “fully disclose to journalists and media professionals how and when AI is used in the workplace” (sponsor’s office, NY Senate press release; confirmed independently in the NewsGuild of New York’s statement and R Street’s opposition commentary). Internal tool usage is not exempt from the transparency requirement — the disclosure obligation runs both outward (to readers) and inward (to employees).
Builder implication: This one primarily affects internal tooling vendors. If you sell AI writing, research, or summarization tools to newsroom operations teams, your customers need to be able to document and disclose those deployments to their reporters. Build audit logs, deployment records, and usage summaries that your customers can produce on demand. If your tool is deployed in a way that is not visible to the employer’s own HR or editorial leadership, that is a liability for the publisher — and a reason for them not to buy your product.
4. Source Shielding and Labor Protections
The bill requires publishers to “establish safeguards to protect journalist sources and confidential materials from being accessed by an artificial intelligence system” (sponsor’s office; NewsGuild of New York). This is a data handling restriction, not a disclosure requirement.
Labor provisions restrict publishers from terminating journalists or reducing pay and benefits due to AI adoption — R Street’s opposition commentary independently confirms the bill would “prohibit AI adoption that results in job displacement, reduced hours, or diminished wages” (R Street Institute).
Builder implication: If you build tools that ingest journalist notes, interview transcripts, or raw documents to generate summaries or story drafts, and those tools route data through an external LLM API, you may be creating legal exposure for your publisher customers. The bill targets the publisher, not the tool vendor directly — but publishers will read their vendor agreements carefully and push compliance obligations upstream via contract.
The practical guidance: if your tool processes journalist source material by sending it to a third-party model API, flag this clearly in your documentation and offer on-premise or private deployment options for sensitive material workflows.
How This Differs from A3411B
These are companion laws aimed at different problems.
| A3411B | NY FAIR News Act | |
|---|---|---|
| Scope | All operators of GenAI systems | News publishers and broadcasters in New York |
| Core requirement | UI notice that outputs may be inaccurate | Authorship label + mandatory human review + newsroom transparency + source shielding |
| Disclosure text | “Outputs may be inaccurate” | “Substantially created by generative AI” |
| Human review | Not required | Required before every AI-assisted publication |
| Internal requirements | None | Newsroom disclosure to staff required |
| Data handling | None | Source material prohibited from external AI processing |
| Effective date | 90 days after signing | 60 days after signing |
| Signed? | Pending | Pending |
A3411B is a product compliance requirement. The FAIR News Act is an editorial and operational compliance requirement — it restructures how AI can be used inside a news organization, not just how the output is labeled.
If you build general-purpose AI products, A3411B is the primary concern. If you build anything specifically for news publishers — article generators, interview summarizers, wire automation, sports brief generators, financial news feeds — the FAIR News Act is the more demanding law.
Status and Timeline
Passed: Senate 53-7 on June 4, 2026; Assembly 130-1 on June 5, 2026, the day the legislature adjourned. (S.8451B full text and action history)
Sponsors: State Sen. Patricia Fahy (D-Albany, 46th District) and Assemblymember Nily Rozic (D-Queens, 25th District).
Governor’s deadline: Hochul has until December 31, 2026, to sign or veto (Transparency Coalition tracking of the 2026 AI bill package). She has not publicly stated a position. Her track record on AI legislation is mixed — she signed the RAISE Act on December 19, 2025, but vetoed the New York Health Information Privacy Act the same month after industry criticism that it was unworkable.
Effective date if signed: 60 days after signing — the bill text specifies “the sixtieth day after it shall have become a law” (S.8451B, Section 4). If signed in July, enforcement could begin September 2026.
Penalties: Violations are civil, enforced by the state Attorney General through injunctive relief: $1,000 for a first offense, $5,000 for each subsequent one (S.8451B, GBL Art. 48 §1803).
Coalition support: NY State AFL-CIO, Writers Guild of America East, WGA West, SAG-AFTRA, Directors Guild of America, NewsGuild of New York, NewsGuild-CWA, CWA District 1, Freelancers Union. (NY Senate press release)
Opposition: R Street Institute and broadcasters argue the human review requirement adds costs that small and local news organizations cannot absorb, potentially accelerating newsroom closures rather than protecting journalists.
The Federal Preemption Variable
The FAIR News Act’s disclosure requirements are compelled speech — the state government requiring publishers to say something about their content. That legal theory is currently under active challenge in federal court.
On December 11, 2025, President Trump signed an executive order, “Ensuring a National Policy Framework for Artificial Intelligence,” directing the Attorney General to establish a DOJ AI Litigation Task Force within 30 days (by January 10, 2026) to challenge state AI laws on interstate-commerce, preemption, and other constitutional grounds. The same order separately directs the FTC Chairman to issue, within 90 days, a policy statement on “the circumstances under which State laws that require alterations to the truthful outputs of AI models are preempted” (executive order text, whitehouse.gov). A comparable compelled-speech theory succeeded, in part, against California’s Age-Appropriate Design Code: the Ninth Circuit found in 2024 that its risk-assessment disclosure requirement likely violates the First Amendment (NetChoice, LLC v. Bonta, 9th Cir. 2024).
This means there is a meaningful probability the FAIR News Act, if signed, faces an immediate legal challenge and possible preliminary injunction before enforcement begins. Builders planning product changes for publisher compliance should factor in this uncertainty: the law may not take effect on schedule, and the compliance requirements could be modified or voided through litigation.
The practical posture: design your compliance features to be toggleable. Build the human review gate, build the disclosure flag, build the audit logging — but don’t hard-code these as mandatory product flows until the law’s status is clear.
What To Build Now
If your products serve news publishers in New York or at national scale, here is what the FAIR News Act implies for your roadmap:
AI authorship provenance: Add a metadata field to every piece of content your tool generates — a boolean or percentage flag indicating how much of the content was AI-generated. Surface this in your export formats, your API responses, and your CMS integrations, so publishers can render the required label without manual tracking.
Human review gate: If you offer a publish or schedule feature, insert a mandatory approval step. The publish action should require a named human editor to confirm before content goes live. Log the reviewer, the timestamp, and the content version they approved.
Audit logging: Add a deployment audit log that shows which AI tools were used, on what content, by which users, and when. News organizations need this to fulfill their internal transparency obligations to staff.
Private data handling tiers: Identify whether your tool routes journalist source material (transcripts, notes, documents) to external LLM APIs. If so, build a private processing option — on-premise deployment, a customer-managed API key with no data retention, or a clear documentation pathway for publishers who need to attest to source-material protections.
Disclosure copy templates: Give your publisher customers ready-made disclosure text and placement guidance. “Substantially created by generative AI” is the required substance; you can help them implement it correctly at article level.
The Broader Pattern
The FAIR News Act is one of an accelerating cluster of state-level AI regulations targeting the news and content space. California, Illinois, Texas, and Washington are drafting or advancing similar measures. Illinois SB 315, which passed the legislature and is awaiting Governor Pritzker’s signature, imposes annual third-party safety audits on AI systems used by news organizations. The EU AI Act’s transparency provisions cover AI-generated news content under Article 52 with machine-readable watermarking requirements beginning August 2026.
The convergent direction: AI content must be labeled, AI editorial decisions must include human checkpoints, and source material used by journalists must be handled with provenance care. Builders who treat these as isolated compliance problems will iterate on each law individually. Builders who treat them as a consistent architectural requirement — provenance metadata, review gating, audit logging, private data tiers — will build once and comply broadly.
The FAIR News Act is not yet law. But the legislative pattern behind it is not going away.
Related: NY A3411B: The GenAI Disclosure Law That Covers Every Builder, Not Just Frontier Developers · New York’s RAISE Act Is Already Signed — the Three-State AI Compliance Stack