Documents unsealed Thursday in the New York Times' copyright suit against OpenAI and Microsoft reveal internal concerns about AI training practices. A 2023 memo by Brent Hecht, a director of applied science, warned that large models "hoovering up" work would be seen as "theft of unprecedented proportions" and described AI as a product destroying its own supply chain. Microsoft stated these views did not represent company positions, noting Hecht was employed to present divergent perspectives.
The filings stem from the late 2023 lawsuit, now joined by eleven other publishers, with Judge Sidney Stein weighing summary judgment motions. Microsoft CEO Satya Nadella testified that paywalled content should be licensed, asserting he would have required retraining had he known OpenAI used such data. Conversely, OpenAI staff discussions included bypassing the Times paywall and acknowledging AI's substitutive nature for publishers. Both companies argue the training constitutes fair use, while plaintiff counsel Steven Lieberman highlighted the contradiction between their legal stance and internal warnings.
The unsealing of these documents exposes a significant divergence between public corporate positioning and internal risk assessments regarding generative AI's impact on content ecosystems. The existence of memos describing AI development as "theft" and a threat to cultural labor suggests that major technology firms were aware of the potential for market disruption and legal liability well before the current wave of litigation intensified. This internal acknowledgment complicates the defense strategy centered on fair use, as it implies an understanding that the technology could degrade the very sources it relies upon for training data.
From a regulatory and institutional adoption perspective, this case highlights the growing tension between intellectual property rights and the scalability of AI models. The testimony from Microsoft’s leadership attempting to distance itself from specific training practices, contrasted with detailed internal warnings, may influence future compliance frameworks and licensing negotiations. Stakeholders should watch how Judge Stein interprets these internal communications when ruling on summary judgment, as the outcome could set precedents for how courts weigh corporate intent and knowledge in copyright disputes involving artificial intelligence.
