Data, Copyright and Licensing

On 27 December 2023 The New York Times sued Microsoft and OpenAI for copyright infringement. Its case went beyond training: it stressed verbatim outputs, retrieval that substituted for visiting its site, and answers wrongly attributed to the Times, a trade-mark claim. The case was consolidated in New York with authors' and other news suits before Judge Stein, who largely denied the motions to dismiss in March–April 2025. A May 2025 order required OpenAI to preserve ChatGPT output logs (reported). Cross-motions for summary judgment followed in September 2026, with a Justice Department statement supporting the defendants and no trial date (reported, verify). In May 2025, while suing OpenAI, the Times signed its first AI licence, with Amazon, reported at $20–25 million a year (WSJ, July 2025, reported), about 1% of its 2024 revenue.

Reddit took the other road first. On 22 February 2024, the day it filed to go public, it announced a partnership with Google that gave Google structured, real-time access to its posts through Reddit's data API for about $60 million a year (Reuters, reported). Its prospectus disclosed data-licensing contracts with an aggregate value of about $203 million over two to three years. OpenAI signed in May 2024. Then on 4 June 2025 Reddit sued Anthropic, alleging that Anthropic's bots had accessed Reddit more than 100,000 times after Anthropic said in July 2024 that it had blocked them (complaint, reported). Its claims rested on contract and unfair competition, not copyright, because its users own the copyright in their posts. In late March 2026 a federal judge sent the case back to California state court, where those claims are tried under state law (reported).

The two cases show the two ways a content owner can make a model developer pay: sue for what was taken, or control access and charge for what is taken next. This chapter explains what the 2025–26 rulings decided, how the licensing market prices content, and why the money and the risk are moving from training to retrieval.