Anthropic reported that it blocked several attempts to use its Claude AI models for potential biological weapons development. The company identified these incidents in a recent report covering the last eight months, noting that users attempted to hide their identities and research objectives.
One specific case involved a researcher investigating the chikungunya virus. Anthropic observed attempts to modify the virus to increase its harmful properties, including plans for practical animal testing in an institution linked to defense research.
The company stated it cannot definitively confirm malicious intent in all cases, as AI queries in this field are also used for legitimate vaccine development. However, Anthropic blocked the associated accounts due to suspicious behavior and notified regulators.
The report also highlights the use of Claude for designing conventional weapons, such as firearms, missiles, bombs, and combat drones. According to Andrew Weber of the Council on Strategic Risks, advanced AI models now pose a dual-use risk for both medicine and bioweapons.
