Anthropic has announced that there were multiple attempts this year to utilize its AI technology for the development of biological weapons. According to a report released by the company, some users—including those from countries where access is prohibited, such as Russia, China, and Iran—attempted to bypass safety measures or conceal their research objectives.

One reported case involved researchers from a non-supported region who spent several weeks using the AI model Claude to plan experiments regarding avian influenza. Anthropic stated that its safety filters restricted this work to the lowest-performance model. Meanwhile, the company emphasized that it cannot determine whether the scientists acted with malicious intent, as the same information could also be used for vaccine development.

Anthropic has already suspended the accounts involved in the reported cases. Additionally, the company stated it detected seven laboratories, including Moonshot and DeepSeek from China, attempting to replicate its technology through a process called "distillation." This is described as part of increasingly sophisticated evasion techniques used to harvest the capabilities of cutting-edge US-made models.

Source: