Publishers Accuse OpenAI of Withholding Evidence in Copyright Lawsuits

The New York Times, Ziff Davis, and 15 other major media organizations have accused OpenAI of deliberately withholding evidence about how it trains its artificial intelligence models, escalating a high-stakes copyright battle. In a motion filed this week, the publishers claim OpenAI 'chose obstruction' instead of cooperating with discovery requests in lawsuits alleging the company illegally used copyrighted articles to train its generative AI systems. This legal move strikes at the heart of a broader debate over whether AI companies can scrape the web freely or must pay for content.

The plaintiffs include some of the most influential names in journalism—from the Times to The Atlantic, Reuters, and others—who jointly argue that OpenAI has failed to produce internal documents detailing the sources and methods used to build its GPT models. The motion asks the court to compel OpenAI to hand over communications, training data logs, and technical specifications that could reveal whether copyrighted works were ingested without permission. The publishers frame this as a matter of basic fairness: if OpenAI wants to argue its use is 'fair use,' it must show exactly what it used.

This conflict is part of a wave of litigation against AI developers. In December 2023, the New York Times filed its own copyright suit against OpenAI and Microsoft, and similar cases have been brought by authors, visual artists, and music labels. The current motion consolidates those efforts, reflecting a coordinated push to force transparency from the industry leader.

Key details: the motion was filed in the Southern District of New York, with Judge Sidney Stein presiding. The 17 publishers specifically cite OpenAI's refusal to provide ‘non-public information about the data sets used to train its large language models.’ OpenAI has responded by calling the accusations unfounded, stating it has already produced over 250,000 documents in related cases. But the publishers counter that those documents skim over training data specifics, leaving critical gaps.

Analysis: This case could set a precedent for AI copyright law. If the court orders OpenAI to open its training-data black box, it may force the company to disclose the balance between copyrighted and public-domain sources—potentially weakening its fair-use defense. Legal observers note that the outcome could reshape how AI firms approach data acquisition, either embracing licensing deals or risking punitive damages.

Outlook: The court is expected to rule on the motion to compel within weeks. A decision favoring the publishers would trigger a new phase of evidence sharing, possibly exposing internal discussions about data sourcing at OpenAI. Meanwhile, other jurisdictions in the EU and UK are watching closely, as similar transparency demands are embedded in upcoming AI regulations.