OpenAI Accuses China’s DeepSeek of Distilling U.S. AI Models in Memo to House Panel

Overview

• OpenAI’s memo alleges DeepSeek employees bypassed access limits by routing through third‑party servers, using programmatic output extraction, and tapping unauthorized resellers to collect model outputs.
• The company warns that replicating capabilities through distillation can strip built‑in safeguards, heightening misuse risks in areas such as biology and chemistry and raising national‑security concerns.
• OpenAI claims DeepSeek’s models display pro‑CCP bias and censor sensitive topics, citing responses on Tiananmen Square and Taiwan independence as examples.
• Reporting notes U.S. investigators are examining DeepSeek’s use of U.S.‑made hardware, with separate accounts suggesting potential access to Nvidia H200 chips remains under review.
• OpenAI argues that open‑weight Chinese models combined with distilled U.S. capabilities intensify cost pressures on proprietary U.S. labs and erode America’s competitive edge in AI.