Anthropic released new threat-intelligence findings on September 10 describing attempts to misuse Claude across cyber operations, surveillance, model extraction and sensitive biological research. Reuters and the Associated Press separately reported on the disclosures, which cover both state-linked and criminal activity.
Reuters reported that Anthropic attributed several campaigns to actors connected with Russia and China, including AI-assisted espionage and large-scale attempts to extract model capabilities. AP separately reported cases in which users sought help with sensitive biological research that Anthropic judged could support harmful dual-use work, alongside cyber and surveillance activity.
The evidence does not mean Claude autonomously initiated every operation described in the report, and AiToolMap does not treat vendor attribution or estimated campaign scale as independently proven beyond the cited reporting. The material is better read as a documented abuse pattern around a widely deployed model family and as evidence of the controls Anthropic says it used to detect or stop that activity.
For AiToolMap, this is a security and governance update rather than a new Claude feature. It matters because the operational risk profile of a general-purpose assistant increasingly includes how providers detect abusive workflows, enforce safeguards and respond when models are used inside larger automated campaigns.