Anthropic Says It Blocked Bioweapons Efforts and Detected Chinese Distillation Attacks

Anthropic has disrupted several attempts to use its Claude models for potentially malicious activity, including research into how to adapt bird flu to a human-transmittable strain with “pandemic potential,” the company said in a report Thursday.

Amid increasing worry about AI misuse and rogue models, the Anthropic report says “biological misuse is one of the most serious risks of frontier AI models.” In one case Anthropic identified, a scientist used Claude to write a grant for research—apparently to be done at a military research institute—into engineering more infectious strains of the debilitating mosquito-borne virus chikungunya.

Anthropic has implemented safeguards where more powerful models refuse to answer potentially dangerous prompts or route the prompt to be answered by less capable models.

Anthropic’s more than 150-page report also said a number of China-based AI model developers including Alibaba, DeepSeek, and Xiaomi used fraudulent accounts to train their models on Claude’s reasoning.

DeepSeek and Moonshot both relayed users who thought they were using the Chinese models to Claude instead, and served Claude’s answers under DeepSeek and Moonshot’s names, Anthropic said. Many of the Chinese developers’ actions resulted in their users’ information being relayed to Anthropic.

The Anthropic report also identified other attempts to use its AI for apparently malicious purposes, including Russian actors using Claude for espionage and propaganda creation, and a new type of misuse where Chinese, Russian, and Yemeni users attempted to write software for weapons development using Claude.