OpenAI and Microsoft move to strike Susman Godfrey-funded market-dilution research from AI copyright case, alleging concealed $100,000 payment

What happened

OpenAI and Microsoft have filed a motion to strike a research paper — "Generative AI floods and dilutes the market for books," a non-peer-reviewed working paper co-authored by Stony Brook University professor Tuhin Chakrabarty and legal scholar Jane Ginsburg — from the sprawling multidistrict copyright litigation against the two companies. OpenAI and Microsoft allege the plaintiffs' law firm, Susman Godfrey LLP (which represents the plaintiff book and news publishers), secretly gave Chakrabarty a $100,000 "gift" to conduct the market-dilution research after failing to produce evidence of harm in discovery, concealed that funding from the court and defense counsel, and backchanneled the paper into the case outside the court-ordered expert disclosure process via a supplemental report by another expert, Michael Lasinski, served on August 2, 2026 — the day before his deposition and after expert-report deadlines had passed.

The same reporting details the plaintiffs' counter-allegations: that OpenAI deleted billions of ChatGPT-generated outputs after the lawsuit began and withheld internal evaluations and training-data search tools, potentially violating court preservation orders; that OpenAI maintained a separate database of 78 million anonymized ChatGPT conversations used to evaluate potential copyright infringement risks; and that it developed an internal detection system called the "Bloom" filter under "Project Giraffe" to identify ChatGPT outputs closely matching training data. Internal Microsoft documents, including statements from executive Brent Hecht, reportedly described AI web scraping as "the largest theft of labor in human history." OpenAI spokesperson Drew Pusateri dismissed the plaintiffs' claims as "blatantly false allegations."

Why it matters to creators

This is the central copyright battle over AI training on creative work — the case news outlets and publishers brought against the two companies that built the most widely used AI products. How the court handles this evidence fight affects what counts as proof that AI outputs harm creators' markets: the contested paper argues generative AI floods and dilutes the market for human-authored books. Separately, the mutual accusations surface facts creators care about: OpenAI apparently built a large internal system to detect when ChatGPT outputs closely matched training data, and plaintiffs allege billions of outputs were deleted after litigation began.

What is not yet known

Whether the court grants the motion to strike and excludes the Chakrabarty paper and Lasinski's testimony; how the court rules on the dueling preservation/discovery allegations; and whether any of the internal training-data evaluation material (the 78-million-conversation database, the "Bloom" filter) ever becomes public evidence.