Moonshot’s Kimi rattled markets. U.S. agencies now say it was trained on American models

U.S. security agencies are accusing six Chinese artificial intelligence companies, including DeepSeek, Alibaba and Moonshot AI, of using American frontier models at industrial scale to help train their own systems.

The NSA, FBI and CISA said Tuesday that these companies extracted billions of tokens across millions of requests from models including Claude, GPT, Gemini and Grok since at least late 2024, often using proxy services and multiple accounts to avoid detection.

The agencies called the practice malicious knowledge distillation and said it likely occurred with the Chinese government’s awareness.

Distillation is a common way to train AI, where developers ask a more powerful model large numbers of questions, collect its answers, and use those outputs to improve another model.

U.S. agencies say the Chinese companies took that much further, sending millions of carefully designed requests to American models to copy their coding, math and reasoning abilities while using multiple accounts and proxy services to avoid detection and usage limits.

Moonshot is the company behind Kimi K3, the open-weight model that topped Anthropic and OpenAI on a widely watched coding benchmark in July and briefly triggered a broader AI selloff.

Chinese AI stocks were hit hardest, with Z.ai falling about 27% and MiniMax about 16%, while semiconductor shares weakened across Asia and bitcoin and other major cryptocurrencies fell alongside them. The pressure lingered into the following session, with South Korea’s Kospi dropping another 3.5%.

CoinDesk reported at the time that Kimi’s performance challenged two big assumptions about AI. One was that building top-tier models would stay enormously expensive, while the other was that the best systems would keep coming from a small group of well-funded U.S. companies.

Moonshot showed that a Chinese lab could release an open-weight model capable of competing with, and on some benchmarks beating, systems from OpenAI and Anthropic without matching the huge spending of the American AI leaders.

That forced investors to ask how durable the U.S. lead really was, and whether the hundreds of billions of dollars being poured into chips and data centers would always produce better models.

Moonshot AI did not respond to a CoinDesk request for comment on the allegations in Asian morning hours.

Read More: Bitcoin faces fresh headwinds as China’s Kimi beats Claude, GPT in coding benchmark