U.S. Agencies Accuse Moonshot AI of Distilling American Models After Kimi Selloff

U.S. security agencies have formally accused Moonshot AI, the developer behind the Kimi model that triggered a significant market selloff earlier this year, of systematically extracting capabilities from leading American artificial intelligence models. The National Security Agency (NSA), Federal Bureau of Investigation (FBI), and Cybersecurity and Infrastructure Security Agency (CISA) issued a joint advisory Tuesday alleging that Moonshot used outputs from Anthropic's Claude and OpenAI's GPT to train its own systems, a practice the agencies term "malicious knowledge distillation."

Market Context

The allegations cast a retrospective light on the market volatility witnessed in July, when Moonshot’s Kimi K3 model topped Anthropic and OpenAI on a widely watched coding benchmark. That performance briefly challenged assumptions about the durability of the U.S. lead in frontier AI, triggering a broader selloff in the sector. During that period, Chinese AI stocks were hit hardest, with Z.ai falling approximately 27% and MiniMax dropping about 16%. Semiconductor shares weakened across Asia, and major cryptocurrencies, including Bitcoin, fell alongside equities. The pressure lingered into the following session, with South Korea’s Kospi index dropping another 3.5%.

Analysis

The core of the agencies' complaint is that Moonshot and five other Chinese firms—including DeepSeek and Alibaba—did not merely observe American progress but actively copied it at an industrial scale. Since at least late 2024, these companies allegedly used millions of carefully designed requests, proxy services, and multiple accounts to evade detection and usage limits. By collecting billions of tokens from models like Claude, GPT, Gemini, and Grok, they replicated coding, math, and reasoning capabilities without matching the massive capital expenditure of their U.S. counterparts.

This development forces investors to reassess the economic moat of U.S. AI leaders. Previously, the market narrative held that building top-tier models would remain enormously expensive and that the best systems would consistently originate from well-funded U.S. companies. Moonshot’s ability to release an open-weight model capable of competing with OpenAI and Anthropic suggested that efficiency and distillation could bypass the need for hundreds of billions of dollars in chips and data centers. If the agencies' allegations hold, the "cost-to-performance" advantage of U.S. labs may be less about proprietary breakthroughs and more about proprietary data access that Chinese firms allegedly circumvented.

The advisory suggests this extraction likely occurred with the awareness of the Chinese government, adding a geopolitical layer to the technical debate. While knowledge distillation is a common training technique, the scale and method described by U.S. agencies—using proxy services to hide the volume of queries—moves the practice from standard industry behavior to what the government characterizes as malicious extraction. Moonshot AI did not respond to requests for comment in Asian morning hours.

Key Numbers

- Z.ai stock fell approximately 27% during the July selloff triggered by Kimi's benchmark performance.

- MiniMax shares dropped about 16% in the same period.

- South Korea’s Kospi index declined an additional 3.5% in the session following the initial AI selloff.

- The agencies allege the extraction of billions of tokens across millions of requests since late 2024.

- Six Chinese AI firms, including Moonshot, DeepSeek, and Alibaba, are named in the advisory.

What to Watch

Traders should monitor potential regulatory responses from U.S. agencies, which may include new restrictions on API access or data usage for Chinese AI firms. The market will also be watching for any counter-statement or legal challenge from Moonshot AI, which has yet to respond. Additionally, upcoming earnings reports from major U.S. AI labs like Anthropic and OpenAI (via parent companies) could reveal if the perceived erosion of their competitive moat impacts guidance or capital expenditure plans.

Key levels to watch include the previous lows established by Z.ai and MiniMax during the July selloff, as renewed fears of Chinese efficiency could test those support zones again. The Kospi index's reaction to any new headlines regarding AI export controls or data sovereignty issues will also serve as a barometer for broader Asian tech sentiment.