US Agencies Raise Concerns Over China’s AI Data Strategies
The NSA, CISA, and FBI have issued a warning about China’s systematic extraction of data from US frontier AI models. Companies such as DeepSeek, Moonshot AI, and Alibaba are reportedly leveraging distillation processes, with possible awareness from the Chinese government, to siphon billions of tokens from models like Claude, GPT, Gemini, and Grok since late 2024.
Impact on US Technological Leadership
This extraction not only enhances Chinese AI capabilities but also poses a significant threat to the US’s technological dominance. Between late 2024 and mid-2025, DeepSeek utilized distilled data, including API-driven tasks and supervised fine-tuning, to advance its R1 and V3 models. Moonshot AI similarly extracted data from Claude Fable 5 and GPT-4o to improve its Kimi-K3 and Kimi-K2 models, respectively.
Detailed Report on Extraction Tactics
The agencies’ report provides an in-depth look at the tactics, techniques, and procedures (TTPs) employed by these Chinese companies, aligning them with the MITRE ATLAS framework. These TTPs outline the process from resource development to execution and impact, describing how these actions could financially harm US frontier model developers and undermine competitive advantages.
Moreover, the report highlights additional techniques not covered by MITRE ATLAS, emphasizing that these actions are strategic and well-resourced. Innovations in evading regional restrictions and exploiting subscriptions are some of the novel TTPs identified.
Recommended Mitigations and Future Outlook
To counter these threats, the agencies recommend a coordinated response across the US AI ecosystem, including cloud providers and infrastructure entities. Suggested actions range from defensive measures like behavioral detection to more aggressive responses against malicious distillation requests.
Implementing differential privacy principles is advised to protect sensitive model information. The report ultimately aims to alert AI stakeholders of the ongoing threat to US technological leadership, which could have broader implications for the economy and national security.
The situation underscores the need for vigilance and proactive measures to safeguard US advancements in AI technology.
