"Existential threat": what AI executives said privately about publishers

Two phrases pulled from newly unsealed court filings have become shorthand for the conflict between AI companies and publishers.

Nick Turley, OpenAI's head of ChatGPT, described publishers as facing an "existential threat" from AI products that were becoming more substitutive. Brent Hecht, a Microsoft director of applied science, wrote that many people would regard AI companies absorbing their work as theft on an unprecedented scale.

The quotes are damaging because they come from inside the companies defending their training practices. They are not judicial findings, and they do not decide whether training is fair use. Their immediate value to publishers may be less dramatic and more commercial: they make it harder to dismiss the economic substitution problem during licensing negotiations.

What was unsealed

Previously redacted portions of summary-judgment filings in the New York Times copyright case became public on September 17. Reuters reported that the publisher plaintiffs cited internal statements from Turley, Hecht, OpenAI president Greg Brockman and Microsoft CEO Satya Nadella.

Turley was quoted describing publishers as facing an "existential threat" from products that are already substitutive and could become more so as they improve. Hecht characterized large-scale absorption of human work as "an astonishing theft of unprecedented proportions." Other reported material addressed the ability of chatbots to answer news questions and replace visits to underlying sources.

The publishers argue that these statements undermine the companies' claim that model training and outputs are transformative and do not substitute for journalism. OpenAI and Microsoft dispute that conclusion. Microsoft told Reuters that Hecht's comment represented one employee's perspective rather than a legal analysis or corporate position.

That distinction is essential. The material appears in advocacy by parties seeking summary judgment. The judge has not adopted the plaintiffs' characterization or resolved the fair-use issue.

Why these statements are different from usual litigation noise

Copyright cases generate competing expert reports, economic models and legal arguments. Internal communications can carry unusual weight because they show what employees believed while products and data practices were being developed.

The statements also address the central commercial question directly. If an AI product can satisfy the same information need as an article, the publisher can argue that the use affects the market for the original work. Market effect is one of the factors courts consider in fair-use analysis, although no single quote resolves that test.

The companies have responses. A description of industry disruption is not necessarily an admission of infringement. Products can alter markets without violating copyright, and an employee's ethical concern does not establish the legal purpose, source or effect of a particular training dataset.

Still, these are harder to treat as outsider speculation. Publishers can point to people inside the AI companies acknowledging substitution and concern about the source ecosystem.

What it changes for publishers negotiating licensing

The filings improve the publisher side's narrative in three ways.

First, they support the argument that content is not an interchangeable raw material. News has particular value because it is current, structured, fact-checked and produced continuously. If AI products are strong at answering news questions, the pipeline that creates reliable new information becomes strategically important.

Second, the reported substitution comments help publishers frame licensing as supply-chain protection, not a request for goodwill. A platform that depends on current reporting has an economic interest in keeping that reporting available, even if the law ultimately permits some training without permission.

Third, the plaintiff base is expanding. More than 550 publications are now represented in a separate coalition action against OpenAI and Microsoft. A larger pool can increase litigation pressure, coordinate evidence and make one-off settlements less likely to settle the broader market.

Negotiators should remain realistic. A sharp internal quote does not create a standard rate card or guarantee that a platform will license every publisher. Scale, exclusivity, archive depth, update frequency and rights clarity will still shape bargaining power.

But the commercial conversation has moved. The question is increasingly not whether publisher content has value to AI systems, but how contribution is measured and whether that value is paid, cited or returned through traffic.

ContentGrip's report on Google's AI contribution pilot shows one possible answer. Google is reportedly testing payments when publisher content significantly helps generate AI answers. Its formula remains opaque, but the pilot accepts the principle that contribution to an answer can have a separate economic value from a click.

The counterweight

The litigation picture is not moving in only one direction.

OpenAI and Microsoft recently fended off part of a separate lawsuit brought by software developers over AI training. Bloomberg Law also described a narrow copyright win that avoided a potential damages threat. Different plaintiffs, works, datasets and legal theories produce different results.

Microsoft has also argued in the publisher litigation that large-scale analysis of Copilot logs showed very few responses substantially matching the plaintiffs' material. That evidence is aimed at separating training from alleged market substitution in outputs.

OpenAI and Microsoft continue to argue that model training is transformative fair use. The US government has supported the view that AI training can be highly transformative. The court still has to weigh purpose, nature, amount used and market effect against the record before it.

Publishers should therefore use the unsealed material as negotiating evidence, not declare victory. A licensing agreement can be commercially rational even when legal liability remains unresolved. Conversely, a partial courtroom win for an AI company does not remove its need for current, reliable content.

What brands should take from it

Most brands are not licensing a news archive or joining a copyright suit. They are still affected by the underlying admission that published material is a critical input to AI products.

That strengthens the case for becoming a source rather than producing generic summaries. Original data, named expertise, documented case studies and clear factual claims give AI systems something distinctive to retrieve and cite. Commodity copy is easier to replace because the same information exists everywhere.

Brands should also review their own rights chain. If an organization expects to license or restrict the use of its content, it needs clear ownership of commissioned research, images, transcripts and contributor work. A valuable archive with ambiguous rights is difficult to monetize.

Finally, content teams should track where their work appears in AI answers and how that exposure affects branded demand. The Wikipedia traffic problem and Google's contribution pilot show the same shift from different directions: published content can influence an answer without producing a visit.

The unsealed quotes do not decide the case. They do make the dependency harder to deny. For publishers at the negotiating table, that is useful leverage even before a judge rules.

This article is produced by ContentGrow. We're building branded media outlets for B2B companies. Interested in learning more? Learn more.