Anthropic Says Claude Used in Possible Bioweapon Research

Anthropic disclosed five notable cases where users leveraged its Claude models to advance biological-research activities with potential dual-use or bioweapon implications — including a state-associated gain-of-function chikungunya grant — and documented attempts to circumvent safeguards via intermediary routing, geographic bypasses, gray-market resellers, and synthetic accounts; Anthropic blocked requests and banned accounts but could not definitively determine malicious intent, underscoring limits of per-prompt safeguards and the need to monitor account behavior and wider context.