Order Is “Clearly Erroneous” Or “Contrary to Law”, Rules Court in OpenAI Case: EDiscovery Case Law

Key Takeaways

• Judge Stein vacated magistrate’s privilege waiver order.
• No waiver found for “non‑use” explanation.
• “Moving target” argument rejected; privilege remained consistent.
• No at‑issue waiver; good‑faith defense not asserted.
• Crime‑fraud exception denied; no probable cause shown.

Summary

U.S. District Judge Sidney H. Stein reversed a November 24, 2025 discovery order that had found OpenAI waived attorney‑client privilege over 2022 communications about deleting two copyrighted datasets, Books1 and Books2. Stein held that OpenAI’s “non‑use” explanation, alleged “moving target” privilege claims, and the assertion that its state of mind was at issue did not constitute a waiver. He also rejected the plaintiffs’ crime‑fraud exception, finding no probable cause that the communications were made to further wrongdoing. The ruling restores full privilege protection for OpenAI’s counsel communications and underscores the high threshold for privilege waivers in copyright litigation.

Pulse Analysis

The New York Southern District’s recent ruling in In re OpenAI, Inc. Copyright Infringement Litigation provides a rare, detailed exposition of attorney‑client privilege in the context of artificial‑intelligence training data. Judge Sidney H. Stein vacated a magistrate’s order that had deemed OpenAI’s communications about the deletion of the Books1 and Books2 datasets privileged‑waived. By emphasizing that a simple factual statement—such as deleting data because it was “not used”—does not reveal legal advice, the court reinforced the protective scope of privilege even when companies discuss internal data‑handling decisions. This clarification arrives at a moment when eDiscovery requests increasingly target the opaque processes behind large‑scale language‑model training.

The opinion also dismantles three common arguments for privilege waiver. First, the so‑called “non‑use” rationale was rejected because it lacked any confidential legal counsel component. Second, the “moving target” theory failed to show strategic inconsistency; OpenAI consistently asserted privilege throughout depositions. Third, the court distinguished a mere denial of willfulness from an affirmative good‑faith defense that would place the state of mind at issue. Moreover, the crime‑fraud exception was denied, as plaintiffs could not demonstrate probable cause that the post‑acquisition communications were intended to conceal wrongdoing. These standards raise the evidentiary bar for future copyright and IP suits involving AI developers.

For AI enterprises, the ruling signals that robust privilege protocols are essential to safeguard internal legal advice from discovery. Companies must document the legal nature of communications and avoid casual disclosures that could be construed as waivers. The decision also serves as a cautionary precedent for litigators seeking to pierce privilege through the crime‑fraud exception; courts will demand concrete evidence of illicit intent. As regulators and content owners intensify scrutiny of data‑scraping practices, the OpenAI case underscores the strategic value of maintaining clear attorney‑client boundaries, potentially shaping risk‑management frameworks across the tech sector.