OpenAI's internal Slack messages could cost it billions in copyright suit (sherwood.news)

Plaintiffs suing OpenAI for copyright infringement have obtained internal Slack messages and emails showing OpenAI discussed deleting a pirated book dataset (linked to LibGen) used in model training, and are now asking a judge to compel production of communications between OpenAI and its lawyers by invoking the crime-fraud exception to attorney-client privilege. They want to know whether counsel advised deletion, which, if proven, could be treated as intentional destruction of evidence and support claims of “willful infringement” — potentially multiplying statutory damages up to $150,000 per work and exposing OpenAI to severe sanctions. U.S. Magistrate Judge Ona Wang has allowed some messages to be withheld while ordering others produced; OpenAI insists it did not waive privilege and that deletion reasons remain privileged legal advice. The dispute underscores a broader industry risk: reliance on pirated corpora (LibGen, “The Pile”) has surfaced in other cases — Meta researchers’ messages and Anthropic’s mixed court outcomes and $1.5B settlement illustrate stakes. Technically and operationally, the case highlights the need for robust data provenance, audit trails, and legal-compliance workflows in training pipelines; informal notes about dataset sourcing or deletion can become discovery fodder with large financial and regulatory consequences. The outcome could reshape how AI labs document training data, engage counsel, and disclose provenance to mitigate catastrophic liability.