Anthropic Report On AI Bioweapon Attempts Divides Scientists
The report puts a specific number and named pathogens behind a general worry about AI-assisted biology, and the split among scientists over how to read those five cases is now the story.
Reporting from 1 source: GIGAZINE.
Anthropic's threat intelligence team published a September 2026 report on misuse of Claude, covering December 2025 to August 2026. It describes five cases in which Claude was used to research pathogens that could cause harm, including three involving virus modification: chikungunya, avian influenza, and Orthopoxvirus work. Anthropic says the scientists involved may have been doing vaccine or drug research, and that it did not assert intent to cause harm. It banned the accounts involved.
Anthropic's report covers malicious use of Claude detected between December 2025 and August 2026, including cyberattacks, influence operations, and fraud. Five cases involved research into pathogens that could cause harm. Three concerned virus modification: a grant application about mutations that would help chikungunya virus spread and evade the immune system, a planned experiment to adapt a highly pathogenic avian influenza strain to mammals, and a grant application about mutations that reduce toxicity in mice for Orthopoxviruses such as monkeypox and smallpox. Anthropic calls the chikungunya work gain-of-function research of extreme concern.
The other two cases involved compiling a list of toxins from public information and computationally redesigning bacterial and viral toxins. Anthropic says the people involved are practicing scientists and that it does not assert they intended harm, but notes they used intermediary servers, VPNs, and disguised accounts to appear to be in unregulated countries. Anthropic banned the accounts; in at least one case the scientists evaded the safeguards.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.