Anthropic Reports Disrupted Claude Misuse in Seven Malicious Harm Areas
September 10, 2026
Anthropic identified and disrupted malicious use of Claude Haiku, Sonnet, and Opus models across cyber, influence, surveillance, biological, and fraud operations between December 2025 and August 2026. The report details how threat actors evolved their methods and how Anthropic updated model safeguards in response to these specific case studies.
HOW THIS AFFECTS YOU
●
builderYou should monitor these evolving misuse patterns to better implement application-level safety guardrails.
●
policyYou can use these case studies to understand evolving patterns in AI-enabled cyber and biological threats.