AI law and the newsroom, September 16–23: OpenAI's own words, a closed copyright door, and a judge's warning on labeling

Between 16 and 23 September, courts and governors decided more about artificial intelligence and journalism than Congress did. A federal appeals court shut down one route copyright owners had been using against AI developers, a judge in Montana doubted a state's deepfake-labeling law, and two governors signed or accelerated rules about who has to disclose what.

Three questions run through the decisions below, and they are the ones this digest will keep asking each week. What may AI companies do with your work? What must you tell your audience when you use AI? Which office, and which document, will ask you about it?

On 17 September, previously redacted passages in the New York Times' copyright case against OpenAI and Microsoft became public, and the news outlets in the case say the quotes gut the companies' fair-use defence. OpenAI's head of ChatGPT, Nick Turley, described publishers as facing an "existential threat" from products that are "largely substitutive" and "will get more and more substitutive as they get better". The filing also quotes OpenAI co-founder Greg Brockman and Microsoft chief executive Satya Nadella. 1

The Times has alleged that OpenAI scraped material from more than 10 million articles, nearly a third of them from the Times itself. 2

The case turns on whether AI training transforms news articles and does not compete with journalism, so a document in which the company's own executive calls the product substitutive goes to the centre of the argument. U.S. District Judge Sidney Stein is weighing the outlets' request for summary judgment. The federal government backed the AI companies on 1 September, telling the court that AI training is "extraordinarily" transformative. 1

On 16 September, the Ninth Circuit affirmed the dismissal of a Digital Millennium Copyright Act claim over GitHub Copilot and OpenAI's Codex, in Doe v. GitHub. The court held that the tools generate new works rather than reproduce stored copies, so leaving out author names and licence terms is not "removing or altering" copyright-management information. 3

Two limits matter for anyone weighing this as a precedent. The court decided only the provision about stripping author and licence information, and it declined to reach, on procedural grounds, whether training on protected material violates that provision. It also tied its reasoning to how these systems work, suggesting that a system which retrieves and republishes stored text could be treated differently. 3

For publishers and authors, the DMCA claim has been one of the faster theories to bring against AI developers. This is the first appeals-court decision on how it applies to generative AI, and it went against the plaintiffs. 3

On 16 September, Governor Gavin Newsom signed SB 1050, which requires a clear disclosure on any video or audio advertisement that uses an AI-generated performer to sell a product or service, and bars continued use of an advertisement found in violation. He signed it at the headquarters of SAG-AFTRA, the union that helped write the language. 4

Two days later he issued an executive order directing the state to accelerate the two AI laws he had signed the previous week: SB 813, which sets up independent verification organizations to assess AI systems, and AB 1405, which creates a registry of AI auditors. The order convenes experts to report within two months on proposals that would require independent third parties to write safety plans for frontier AI companies, verify their filed transparency reports and risk assessments, and build an emergency shutoff for frontier models. 5

Read together, the two moves describe where California is heading for news organisations: disclosure duties over synthetic media on the advertising and creative side, and a compliance industry of verifiers, auditors and filings on the developer side. Both create paper that reporters can ask for.

On 21 September, Governor Kathy Hochul announced the first implementation steps for New York's RAISE Act. Large frontier AI developers will be directed from November to register with the state and prepare for obligations that begin in January 2027: publishing safety and transparency frameworks, reporting critical safety incidents within 72 hours, and filing quarterly assessments of catastrophic risk. 6

The filings go to a new office inside the Department of Financial Services — the Office of Digital Innovation, Governance, Integrity and Trust. Members of the public may report suspected safety incidents to it, and the office will put out an annual public report on the incidents it received, what it observed about frontier model safety, and any changes it wants to the law. 6

For a reporter, that report and those quarterly risk assessments become a new source of documents about the most powerful models in use, filed with a state agency rather than released at a company's discretion.

On 18 September, U.S. District Judge Susan Watters ruled that a Montana law regulating AI-altered images in election mailers likely violates the First Amendment. The case started with mailers sent before June's primary that used AI to change images of candidates. Watters wrote that the law treats "identically deceptive media differently based solely on whether its message is weaponized or deployed to elevate a candidate", which makes it viewpoint-based, and called it presumptively unconstitutional. Her ruling covers only the plaintiffs in the case and stops short of striking the law down. 78

The objection reaches further than one state's election rules, because AI-labeling mandates are being tested on free-speech grounds. New York's FAIR News Act, which would require news organisations to label news content "substantially composed, authored, or otherwise created" by AI, drew a similar warning in July from Gabe Rottman of the Reporters Committee for Freedom of the Press, who argued that commercial disclosure rules and editorial content sit in different places under the First Amendment. The bill passed both chambers in June 2026 and was awaiting Governor Hochul's signature as of 23 July reporting. 9

On 19 September, President Donald Trump said he will form an "AI Force" and appoint an artificial intelligence tsar. In a social media post he said his administration "will not in any way hinder or stifle the growth" of the industry and pushed back on calls to slow development. He offered no details and no timeline. 10

The announcement arrived in the same week that two states moved into implementation of binding AI rules, and it changes nothing a newsroom has to file or disclose.

Obligations that are already in force, or already dated, for anyone publishing or building with AI:

California's legislature has adjourned, and Governor Newsom has until 30 September to sign or veto the measures still on his desk. Among them are amendments to the California AI Transparency Act that would extend provenance, verification and disclosure duties for AI-generated or AI-altered content. California has no pocket veto, so anything he leaves unsigned becomes law anyway. 11