OpenAI Disrupts Chinese-Linked Model Distillation Campaign

OpenAI on September 30 disclosed it disrupted a coordinated campaign by a Chinese-linked AI lab to extract protected reasoning outputs through adversarial distillation. The attackers, unable to break encryption directly, copied encrypted reasoning outputs from one conversation and asked a separate model to decrypt and transcribe them. The campaign began July 1 at low volume, then spiked July 24-25 with 16,000 requests from over 4,000 users; related activity spanned a cluster of more than 15,000 users. OpenAI said it fully disrupted the activity by July 28, closing the replay pathway, adding output-holding checks for streamed reasoning, banning involved accounts, and coordinating with third-party services used by the operators.

The September disclosure fits into a longer policy arc. In February 2026, OpenAI sent a memo to US lawmakers alleging that DeepSeek used distillation to transfer learnings from OpenAI's systems and accused Chinese large language models of cutting corners on safe training and deployment practices. The White House OSTP followed in April 2026, with Director Michael Kratsios signing NSTM-4 directing federal agencies to share intelligence with AI companies and explore accountability measures. A CNAS paper from June 2026 noted that the Deterring American AI Model Theft Act of 2026 (H.R. 8283) had been introduced to target actors linked to countries of concern.

Nvidia CEO Jensen Huang, in a CNBC interview published September 28, took a contrasting position, characterizing distillation as routine competitive behavior.