US Agencies Accuse Chinese AI Firms of Extracting Data From American Models

The NSA, FBI and CISA have accused six Chinese artificial intelligence companies, including DeepSeek, Alibaba and Moonshot AI, of systematically pulling billions of tokens from leading American AI models to train their own systems. According to the agencies, the companies extracted data from models including Claude, GPT, Gemini and Grok since at least late 2024, often relying on proxy services and multiple accounts to avoid detection.

Practice Described as Malicious Knowledge Distillation

Officials said the technique went well beyond standard distillation methods used across the AI industry, involving millions of carefully crafted requests designed to replicate coding, math and reasoning capabilities from US systems. The agencies suggested the activity likely occurred with awareness from the Chinese government.

Kimi’s Earlier Market Impact Adds Context

Moonshot is the company behind Kimi K3, an open-weight model that outperformed competitors on a major coding benchmark in July, triggering a broader selloff across AI related stocks and cryptocurrencies at the time. Chinese AI shares dropped sharply, semiconductor stocks weakened across Asia, and bitcoin fell alongside other major tokens. The episode had raised questions about whether the cost advantages long associated with leading US AI labs would hold, given a Chinese lab’s ability to produce a competitive model without matching that level of spending.