OpenAI just lost a major discovery fight, one that could leave the company on the hook for billions of dollars in damages after allegedly pirating huge datasets of copyrighted books.
In the latest twist in the escalating legal battle, a federal judge ordered OpenAI to hand over internal documents explaining why it deleted two massive book datasets, known internally as “Books 1” and “Books 2.” These datasets allegedly contained thousands of pirated works scraped from shadow libraries, and the court’s ruling now clears the way for plaintiffs to probe whether OpenAI knowingly infringed.
If the authors suing the company can prove OpenAI acted “willfully,” the company could be liable for up to $150,000 per infringing work, a number that quickly scales into the billions given the volume of books involved.
A Discovery Meltdown Inside OpenAI
The trouble began when authors and publishers obtained Slack messages showing employees discussing a coordinated effort to erase the pirated book datasets. OpenAI claimed the materials were deleted simply because they were “not being used,” but then tried to walk that statement back, insisting the deletion rationale was protected by attorney-client privilege.
Judge Ona Wang wasn’t having it.
In a sharply worded ruling, she said OpenAI had “made a moving target of its privilege assertions” and ordered the company to turn over documents and sit for depositions. That includes Slack conversations from internal channels ominously named “project-clear” and “excise-libgen.”
In short: OpenAI’s attempt to shield the communications may have backfired.
The ruling gives authors a clearer path to arguing that OpenAI engaged in willful infringement, the most serious violation under copyright law. If the court finds OpenAI knowingly downloaded and stored pirated books, regardless of whether they were used in training, the financial damages could be enormous.
The theory has already gained traction. In a parallel case against Anthropic, a federal judge reaffirmed that illegally downloading copyrighted books is a standalone infringement, even if the company later bought legitimate copies. Anthropic settled that case for $1.5 billion.
Now OpenAI is staring down a similar threat, and possibly a much larger one.
The Stakes: Billions and the Stability of OpenAI’s Defense
If the court concludes that OpenAI destroyed evidence to hide wrongdoing, future juries could be instructed to assume the deleted materials would have been harmful to the company’s case. That would make defending against the infringement allegations nearly impossible.
It could also open OpenAI up to the maximum statutory damages, again, $150,000 per work, for potentially tens of thousands of books.
OpenAI continues to insist it did nothing wrong and is now trying to pause enforcement of the discovery order. But after this ruling, the momentum has clearly shifted toward the authors.
Source: The Hollywood Reporter.