Anthropic reports attempts to misuse AI for biological weapons research
Anthropic said it blocked several requests to use its AI models for work that could aid the development of biological weapons and called for tighter access controls.
Anthropic disclosed that it had intervened in multiple attempts to employ its artificial-intelligence systems, including the Claude chatbot, for research that could facilitate the creation of biological weapons. The company’s threat-intelligence team compiled a report covering the past eight months, noting that while it could not always confirm malicious intent, it chose to err on the side of caution.
One incident in May involved a scientist requesting assistance with a grant proposal to engineer more harmful mutations of the chikungunya virus, a mosquito-borne pathogen. Anthropic identified that the request appeared to originate from a military research institute and subsequently banned the associated accounts.
"What we don’t know is if the research was meant to weaponised," said Jacob Klein, Anthropic’s head of threat intelligence.
The report also highlighted inquiries to use Claude for missile development and espionage. Andrew Weber of the Council on Strategic Risks, who reviewed the findings, said access to such models should be limited to trusted researchers.
Anthropic noted that earlier evaluations indicated its older models were unlikely to aid dangerous biological work, but newer versions are capable of performing complex scientific tasks. The company’s actions come amid growing debate over AI safety and calls for stricter controls.