Anthropic says Claude was used in weapons development, surveillance and cyberattacks

Anthropic has disclosed that its Claude models were used in weapons development, surveillance and cyber operations, according to a threat intelligence report the company published this week and covered by Impakter. The cases span military systems, offensive cyber work and mass surveillance.

According to the report, China-based actors used Claude to develop air-defense and anti-torpedo systems, while Russia-based actors used the models to build software for autonomous attack-drone swarms.

In cyber operations, Chinese-speaking operators used Claude against roughly 50 organizations, including foreign government networks. Anthropic said AI-driven workflows helped them identify vulnerabilities, develop exploits and carry out intrusions with limited human supervision.

The report also identified five cases involving attempts to use Claude for research that could support biological weapons development, the category with the highest potential for harm.

On surveillance, Anthropic reported AI-assisted activity targeting Uyghurs, dissidents and political groups. In Mali, Claude helped build a system capable of monitoring data linked to about 25 million SIM cards across the country's three mobile operators.

The significance is less about any single case than about how low the barrier has become. When one frontier model can be pointed at air-defense systems, drone-swarm software and exploit development, keeping it out of military, intelligence and criminal hands after release becomes a standing operational fight rather than a one-time policy decision.

Publishing the cases also serves Anthropic's regulatory argument: concrete evidence of misuse is the strongest material for rules that tie model access to capability evaluations and use restrictions. What to watch next is how the flagged accounts and uses are handled, whether detection methods are shared, and whether other labs disclose at the same level of detail.