Anthropic released a 154-page threat intelligence report detailing five specific cases where users attempted to utilize its Claude AI models for biological weapons research. The company stated that it blocked every identified operation, noting that one user managed to draft a smallpox-related grant application in one hour using the Opus model.
The report highlights a 30-day investigation into state-linked institutions, which uncovered 35 separate research efforts. While Anthropic blocked these attempts, it noted that some users bypassed restrictions by utilizing fallback mechanisms to rival AI models. The company emphasized that it does not assert malicious intent, as the same scientific knowledge can be used for both vaccine development and weapon creation.
Beyond biological research, the report covers misuse of Claude for cyberattacks, surveillance, and influence operations. Anthropic confirmed it has banned the accounts associated with these activities. The company also noted that its filters successfully redirected a bird flu researcher to weaker models, characterizing the assistance provided by the AI as primarily clerical.
The findings arrive amid increasing regulatory scrutiny in the United States. In July, representatives Ted Lieu and Nathaniel Moran introduced the AI Kill Switch Act, which would require developers to maintain the capability to shut down their systems. Anthropic previously faced challenges in April when a Discord group accessed its restricted model on the first day of availability.
