Anthropic Report Details State Actors Using Claude for Espionage, Missile Software and Mass Surveillance
Ourtake
Anthropic concedes its own classifiers cannot reliably separate dual-use intent from harmful intent, so filtering alone is not a complete defense.
What happened
Anthropic's threat report describes state-linked actors using Claude for espionage malware that rewrote itself to evade antivirus, missile guidance software, an autonomous drone swarm design, and a mobile surveillance platform in Mali. It also describes Chinese AI labs running large distillation campaigns against Claude, some relayed through their customers' requests.