OpenAI Emails Show Executives Viewed Publishers as Existential Threat

Newly unsealed documents in the lawsuit between The New York Times and OpenAI reveal that tech executives viewed the publishing industry as a direct competitor rather than a source of raw material. The files, made public last week, contain internal communications where leaders at OpenAI and Microsoft discuss the economic implications of their products for news organizations.

These disclosures provide the clearest look yet into the strategic thinking behind AI training data. They suggest that the companies understood their models could replace the need for users to visit news sites, a fact that complicates their legal defense that such use is fair and non-infringing.

Executives acknowledged substitution of news sources

Internal emails show a stark awareness of how AI chatbots alter user behavior. Microsoft CEO Satya Nadella testified that conversing with these tools substitutes the need to go to the underlying source, effectively providing information directly on the AI platform. This admission undermines the argument that AI uses news content for a transformative purpose rather than as a direct replacement.

OpenAI leadership also documented the high performance of their models in generating news-like text. Nick Turley, head of ChatGPT, described the products as largely substitutive for publishers. Greg Brockman, OpenAI’s president, noted that the models are particularly good at predicting the text of news articles. These statements highlight that the companies knew their tools could mimic journalism closely enough to reduce traffic to original sources.

Fair use defense faces significant legal hurdles

Legal experts indicate that these documents severely weaken the fair use defense. A key factor in fair use analysis is the market impact on the original work. The filings show that a licensing market for this content already exists, with major AI players signing deals with publishers for similar uses. By refusing to pay for access while acknowledging the substitutive nature of their products, OpenAI and Microsoft face a difficult legal position.

The existence of paid licensing agreements elsewhere in the industry suggests that the content has a clear market price. As reported by Digiday, this conduct establishes that the defendants were aware of the value of the data they used. Their refusal to pay the plaintiffs for this access is now central to the lawsuit, as it contradicts the notion that the use was harmless or transformative.

Data shows sharp drop in publisher clicks

Quantitative data included in the filings supports the claim of market harm. Microsoft recorded significant declines in click-through rates for news domains when comparing traditional search results with its Copilot chatbot. The drop reached as high as 93% for certain outlets. This metric demonstrates a tangible shift in user behavior, where AI answers replace the need to visit the original news site.

These figures provide concrete evidence of the economic impact on publishers. They show that the substitution is not just theoretical but measurable in lost traffic. For publishers, this data strengthens their argument that they are suffering direct financial harm from the unauthorized use of their content in AI training and output generation.