Anthropic Blocks AI Misuse for Bioweapons, Cyberattacks

Anthropic said it had thwarted several attempts over the past eight months to misuse its Claude models for developing biological weapons, military software and cyberespionage, according to the company's latest threat intelligence report.

The report documented five cases in which researchers used Claude in ways that could have aided the development of biological weapons. In one case, a researcher based in a region not supported by Anthropic used the model to plan experiments on adapting avian influenza to mammals. Anthropic explained it had blocked the relevant accounts and strengthened its security measures.

The report also detailed the use of Claude by actors in Russia, China and Yemen to develop software for missiles, armed drones and other conventional weapons.

One hacking group used AI in nearly every phase of cyberattacks against Ukrainian government, military and diplomatic targets, Anthropic noted. The group's methods matched those of Midnight Blizzard, a Russian group that the US government links to Russia's foreign intelligence service (SVR).

Anthropic also mentioned it had thwarted attempts by seven Chinese technology labs, including Alibaba and DeepSeek, to acquire Claude's capabilities in order to train their own models.

(Reuters, max)