Anthropic reported that its Claude AI models successfully blocked several attempts by state-sponsored actors to conduct research related to bioweapons. The company identified five specific cases involving the development of more virulent viruses and toxic compounds.
In these instances, users employed anonymization techniques, including virtual private servers and U.S.-based proxies, to bypass regional restrictions. Anthropic maintains blocks on access to its services from countries including China, Russia, Iran, and North Korea.
The investigated cases included requests to enhance the chikungunya virus, study the adaptation of avian flu to mammals, and analyze orthopoxviruses. Researchers also attempted to use the platform to optimize toxic peptides and bacterial toxin subunits.
Anthropic discovered that these inquiries often originated from military-linked institutions or state-sponsored programs. The company banned the associated accounts and shared the findings with government authorities.
