Federal agents from the NSA, FBI, and CISA have identified a systematic effort by Beijing-linked companies to copy the internal logic and capabilities of American artificial intelligence. These industrial-scale knowledge distillation campaigns involve millions of requests and billions of tokens to extract proprietary functionalities from models including Google Gemini, OpenAI's ChatGPT, Anthropic's Claude, and xAI's Grok since late 2024.
The joint advisory names six Chinese firms, including Alibaba, DeepSeek, and Moonshot AI, as actors using these methods as the core of their AI development strategy. By mid-2026, Z AI had distilled data from GPT-5.5 and Claude Opus 4.8 to create chain-of-thought reasoning capabilities, a process US officials warn could bolster China's military and cyber power weeks before a scheduled meeting between Donald Trump and Xi Jinping.
Key sources
- SOURCE@cisagov“joint advisory with @NSAGov & @FBI details the threat”x.com
- SUPPORT@cyberscoopnews“form the core—not merely a supplement—of their AI development strategy”x.com
- SUPPORT@kyleichan“distilled billions of tokens of GPT-5.5 data and Claude Opus 4.8 data to develop the CoT reasoning capabilities of its model”x.com
- SUPPORT@cointelegraph“weeks before a planned Trump-Xi meeting”x.com
- SOURCE@nsacyber“highlighting AI knowledge distillation, TTPs used, and recommended mitigations”x.com
- SOURCE@deitaone“Beijing called the allegations “unfounded” and urged Washington to pursue greater AI cooperation instead”x.com
- SUPPORT@financialjuice“china firmly opposes this. US allegations are baseless”x.com
- SUPPORT@financialjuice“Distillation is a common practice among AI models globally, including by US firms”x.com