Chinese military researchers and tech giants caught using Claude — US frontier model coded 16 air-defense suppression tools targeting Taiwan, drafted anti-torpedo specs, and fed 151 million training queries to Alibaba
Anthropic accuses China-linked actors of using its Claude models of weapon development, intelligence activities, distilling AI models.
China-linked individuals and organizations, according to Anthropic's September 2026 threat report, have utilized Claude for various programs, including two military initiatives, two surveillance operations, and one focused on understanding Claude's capabilities. One Chinese defense contractor employed Claude to draft a fire-control specification for an anti-torpedo system, test it against U.S. Navy's systems, and prepare a 200+ page proposal for a potential client.
The contractor claimed to be a U.S. defense sector OEM, but Anthropic suspects a connection to the People's Liberation Army Navy. Another Chinese researcher used Claude to develop 16 software modules for electronic warfare and suppressing enemy air defenses, analyzing radars, SAM sites, command posts, and communications nodes, with some scenarios featuring 12 targets in Taiwan.
It is notable that two Chinese military-related projects relied on Anthropic's Claude, despite Chinese-language prompts and other evidence pointing to PLA involvement. It is more likely that Claude was chosen for its proficiency in coding, reasoning, and agentic tasks, or due to the extensive English-language training of U.S. frontier models, which may provide more knowledge about American military technologies and systems.
Additionally, Anthropic discovered China-linked surveillance operations targeting Uyghurs outside of China, where a Chinese government-linked actor used Claude to infiltrate Uyghur armed groups in Syria, surveil Uyghur diaspora activists and media, and collect data from over a hundred WhatsApp groups and Telegram channels. The actor also created deceptive approaches in local dialects and evaluated the credibility of recruitment messages.
The most concerning aspect is that Chinese entities are stealing from Anthropic by employing distillation techniques to extract Claude's reasoning and capabilities, training their own models in the process. Alibaba's campaign, involving over 151 million Claude interactions between May and July 2026, resulted in Qwen 3.x being trained for reasoning, coding, agentic software engineering, kernel development, and long-horizon tasks.
Other Chinese AI developers, including DeepSeek, Xiaomi, Zhipu/Z.ai, and others, are also accused of similar industrial-scale distillation campaigns.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
