US agencies say Chinese AI firms copied American models, targeting Moonshot's Kimi
US agencies allege Chinese AI firms copied US models
The National Security Agency, Federal Bureau of Investigation and Cybersecurity and Infrastructure Security Agency released a joint statement on Tuesday accusing six Chinese artificial‑intelligence companies, including Moonshot AI, of extracting large amounts of data from leading American models such as Claude, GPT, Gemini and Grok.
Key points
- Companies allegedly used billions of tokens across millions of requests.
- Proxy services and multiple accounts were employed to avoid detection and usage limits.
- Moonshot’s Kimi K3 model topped Anthropic and OpenAI on a coding benchmark in July, sparking a brief market sell‑off.
- Chinese AI stocks fell (Z.ai ~27%, MiniMax ~16%) and Bitcoin and other cryptocurrencies dropped alongside them.
Agency findings
The agencies described the practice as “malicious knowledge distillation,” where outputs from powerful models are collected and used to improve another model. They said the activity has been ongoing since at least late 2024 and likely occurred with the knowledge of the Chinese government.
Market reaction
After the July benchmark result, Chinese AI shares fell sharply, semiconductor stocks across Asia weakened, and major cryptocurrencies such as Bitcoin slipped. The pressure continued into the next trading day, with South Korea’s KOSPI index dropping another 3.5%.
Why it matters
The allegations raise concerns about intellectual‑property theft in the fast‑moving AI sector and suggest that competitive advantages may be eroded if foreign firms can replicate U.S. capabilities at lower cost.
Unclear aspects
Moonshot AI did not respond to a request for comment. The agencies have not provided technical proof of the alleged data extraction, so the claims remain unverified outside the official statements.
