Anthropic has published a new threat-intelligence report detailing real-world attempts to misuse Claude across seven harm areas, based on activity the company says it disrupted between December 2025 and August 2026.
The cases span cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional-weapons work and unauthorized model distillation. Anthropic says it banned accounts or organizations involved in the cases and used the findings to strengthen its safeguards.
The report is notable for the level of operational detail. Anthropic describes Claude being used as an engineering or analysis component in surveillance systems, cyber workflows and other high-risk projects. It also says some actors tried to fragment harmful work into smaller requests, an approach that could make malicious intent harder for safeguards to identify consistently.
Independent reporting from Reuters corroborates the publication and its broad findings, including attempts to use Claude in weapons-related research, cyber operations and surveillance. The individual case descriptions remain Anthropic’s own investigative findings rather than independently reproduced incidents.
For AiToolMap, this is a material Claude safety update rather than a capability launch. It provides new evidence about real-world misuse patterns and the safeguards Anthropic says it has deployed in response; it does not establish that every described project was successfully completed or operationalized.