Executive Summary

New York's legislature passed the FAIR News Act through both chambers. The law does not stop at asking outlets to disclose at the top of the page that a piece was made by AI. Any AI-generated article, audio, image, or video must be reviewed by a person with editorial authority before it is published, and reporters' sources and confidential materials must be technically walled off so AI systems cannot touch them. This piece looks at what those three layered duties define, and at how they diverge from the labeling rule of the EU AI Act, whose enforcement is already close.

The point is that the place regulation reaches has moved. Where the EU AI Act aimed at the output stage — telling consumers "this was made by AI" — New York's law steps inside the workflow that produces that output. It nails down where a human must intervene, and it designates the data zones AI must not enter. Labor provisions ride along too, barring the use of AI to replace staff or cut wages.

Governor Kathy Hochul's signature is still pending, and the law would take effect 60 days after she signs. The New York News Publishers Association is pushing back, arguing that forcing newsrooms to carry specific disclosures violates the First Amendment's free-speech protections. In the gap between passage and effect, it is worth reading the signal: law is beginning to dictate, directly, who may lay hands on data.

60 days

from signature to effect

Awaiting Gov. Hochul's signature

$5,000

per repeat violation

First violation is $1,000

76%

of Americans worried about AI stealing news

The public mood behind the bill

3,500+

U.S. newspapers closed since 2005

Newsroom staff down ~2/3 in a decade

The New York Times newsroom in 1942 — copy editors reviewing wire dispatches at the telegraph desk
▲ The New York Times newsroom, 1942 — copy editors review wire dispatches. Eight decades later, New York State makes human review a legal requirement in the AI era. | Source: Wikimedia Commons (Marjory Collins / U.S. Farm Security Administration, 1942 — Public Domain)
1

Three Duties That Go Past the Label

The NY FAIR News Act (formally the New York Fundamental Artificial Intelligence Requirements in News Act) was introduced by State Senator Patricia Fahy and Assembly Member Nily Rozic. Rather than papering over the problems AI brings into the newsroom with a single disclosure, it intervenes at three separate points, each in a different way. Take them one at a time.

NY FAIR News Act — Three-Layer Regulatory Structure ① Disclosure duty AI-generated articles, audio, images, video — conspicuous notice at top (shared with EU AI Act Article 50) EU AI Act + NY ② Human editorial review gate Review by person with editorial authority required before publication — human checkpoint in workflow NY new ③ Sources & Confidential Materials — AI access blocked Reporters' sources and confidential data technically walled off from AI system access (§ 79-h) NY new Output → Process → Data access — the layer regulation targets moves step by step
▲ NY FAIR News Act — three-layer structure. Where the EU AI Act stopped at output labeling, New York reaches into the process and data-access layers. | Pebblous original diagram
Duty 1 · Disclosure

1.1Flag AI-made content conspicuously

Content substantially written or generated by AI must carry a conspicuous notice of that fact at the top of the page. For audio content, the disclosure must be spoken at the start of the broadcast. The scope reaches beyond articles to audio, images, and video. Up to here, it is not far from the labeling idea in the EU AI Act.

Duty 2 · Human editorial review

1.2A person must review before publication

Here is the law's first point of difference. All AI-generated news content must be reviewed by a person with editorial authority before it is published. The statute (General Business Law § 1154) states plainly that such content may be published only after review by a person holding editorial authority.

A label is a tag that attaches to the output; the review duty governs the workflow before the output reaches the world. It means that even if AI writes the draft, a human is mandatory at the final gate where private text becomes public information.

Duty 3 · Source protection

1.3Wall sources and confidential materials off from AI

The second point of difference is on the data side. Employers must put safeguards in place so that reporters' sources and confidential materials are not accessible to AI systems (a new provision added to Civil Rights Law § 79-h). This holds for information gathered by any means, including location tracking or surveillance.

In practice, it is a demand to technically block the act of feeding confidential whistleblower documents or raw interview materials into an outside machine-learning model. In effect, it designates certain categories of data as zones AI cannot enter.

Two more layers attach here. One is worker disclosure. News media employers must fully disclose to employees how and when they use AI tools, and which systems they use for what purpose. The demand is to make AI's points of intervention transparent not only to readers but to the people working alongside it inside the newsroom. The other is labor protection. The law bars using AI to replace staff, cut hours or wages, or weaken collective bargaining agreements. To train AI on an employee's work, an outlet must obtain prior consent and negotiate compensation (§ 1155). Penalties run $1,000 for a first violation and $5,000 per violation thereafter. That is the backdrop against which labor and creative groups — the New York State AFL-CIO, the Writers Guild (WGA East and West), SAG-AFTRA, and the NewsGuild — lined up behind the bill.

2

How It Differs from the EU AI Act

In our earlier piece on the EU AI Act's August 2 transparency obligations, we saw that Article 50 requires chatbots to disclose that they are AI and deepfakes to be marked as "artificially generated." It is label-centric regulation, fixed to the output, protecting a consumer's right to know what they are looking at.

The NY FAIR News Act takes that label as its starting point and asks for two more things. It mandates human review before the output exists, and it cuts off AI access to specific data at the source. The layer regulation targets has shifted from the output to the process and to data access.

Dimension EU AI Act Article 50 NY FAIR News Act
Regulated layer Output (label) Output + process + data access
AI content "Artificially generated" marking Top-of-page disclosure + human review before publication
Human intervention None (label only) Review by editorial authority required before publication
Sources & confidential materials Not addressed AI access blocked at the source
Labor protection Not addressed Bars AI-driven staff replacement and wage cuts

Lay the two laws over each other and regulation's direction of travel comes into view. It began as "put a mark on what AI made," and now it defines "which data AI may touch and when a human must step in." The label protects the consumer; review and access control aim at the very way information gets made.

3

Data Governance's New Front Line

From the vantage of someone who has worked in data governance, the source-protection clause reads like a familiar idea recast in the language of statute: data classification and access control. The sentence "a reporter's sources are a zone AI cannot enter" amounts, in the end, to assigning an access tier to a particular category of data and blocking any system that crosses that boundary.

Data Access Boundary — Zones AI Cannot Enter AI System (news generation) access control boundary General News Data ✓ AI access allowed Sources & Confidential ⊘ AI access blocked (§ 79-h) Data classification and access control — organizational discretion becoming regulatory mandate
▲ Data access boundary — separation of AI-accessible zones from walled-off zones. Reporters' sources and confidential materials are designated as areas AI cannot enter. | Pebblous original diagram

The human-review duty reads along the same grain. It plants a human gate at the point where AI output converts into public information. That is no different from the design of inserting a mandatory verification step into a data pipeline. Access rights and verification procedures that organizations used to design at their own discretion are now moving into requirements that regulation spells out.

There is no reason this pattern stays in the newsroom. In high-sensitivity domains — patient records, legal documents, financial transaction data — the same question holds: which AI may access this data, and at what stage. Drawing a clear line around who may touch which data is settling in as both a design principle for enterprise data architecture and a condition of regulatory compliance.

Pebblous note. Tracing where data came from and managing its access boundaries overlaps with what Pebblous has worked on in AI-Ready Data. Less that regulation created a new market, and more that the old practice of classifying data and controlling access is now being rewritten in the language of law.

4

Objections and Open Variables

The law did not settle into place without friction. The New York News Publishers Association (NYNPA) sees the mandate to carry specific disclosures in the newsroom as compelled speech, which the First Amendment forbids. Citing precedents such as Wooley v. Maynard and Miami Herald v. Tornillo, the core of the objection is that the state cannot dictate to the press what to say or how to say it.

New York State Capitol building in Albany — where the NY FAIR News Act passed both chambers of the legislature
▲ The New York State Capitol in Albany — where the NY FAIR News Act cleared both the Senate and the Assembly. Governor Hochul's signature is all that remains. | Source: Wikimedia Commons (David Brossard, 2021, CC BY-SA 2.0)

Groups like the R Street Institute worry from a different angle. Adding a compliance burden to a journalism industry already in financial distress could hasten the closure of small and mid-sized outlets. With the top 46 news sites having seen revenue fall 56% over the past decade and more than 3,500 newspapers shuttered, the argument is that the cost of regulation topples the weakest first.

Two variables remain. One is whether Governor Hochul signs. She has not yet, and even a signature leaves 60 days before the law takes effect. The other is the legal fight expected to follow. The First Amendment question is likely to become litigation at the enforcement stage. Passage is not the same as certainty, and the flow is worth watching with that in mind.

R

References

Official Sources

Industry Coverage