Anthropic has published a threat-intelligence report describing cyber operations in which actors used Claude in ways that went beyond ordinary chatbot assistance. The company says it disrupted activity involving suspected state-backed groups, financially motivated criminals and politically motivated individuals between December 2025 and August 2026.
One case, designated GTG-20006, is assessed as consistent with Russian state-nexus espionage and targeted government, diplomatic and defense-related interests. Anthropic says the operation used AI-assisted workflows across infrastructure acquisition, phishing, persistence, command-and-control and data exfiltration.
The report’s most consequential defensive finding is the use of automated feedback loops: when monitoring agents detected that malware had been caught by a security product, they allegedly modified and rebuilt the toolkit until it was no longer detected. That pattern puts pressure on static detections and reinforces the value of behavior-based monitoring, rapid detection engineering and controls around identity, phishing and command-and-control activity.
Anthropic cautions that the examples are notable cases rather than a measure of typical misuse. Still, its report argues that AI is increasingly being used to orchestrate parts of the attack chain, reducing the labor and specialist knowledge required for some operations.
Source: Anthropic, “Detecting and countering misuse of AI: September 2026”.