How Anthropic says Claude was used for weapons, spying and cyber operations

Anthropic, an AI company, warned in a recent report that its Claude AI models have been misused for various malicious purposes, including weapons development, cyber operations, surveillance, and fraud. Several threat actors have exploited Claude's capabilities to create an electronic-warfare and air-defense suppression software suite, an anti-torpedo system for China's navy, and specifications for a guided rocket and ballistic missile with hypersonic glide vehicle.

Chinese actors have also used Claude to research high-power microwave weapons, identify components for reverse-engineering, and draft restricted briefings for senior officials. In Yemen, a cell of threat actors utilized Claude to develop software for a guided rocket and a missile variant with hypersonic glide capability, though Anthropic has no evidence they successfully deployed the weapon.

Russian entities allegedly used Claude to create software for autonomous FPV attack drones, while Chinese-based procurement managers sought intermediaries in China and Hong Kong to acquire goods with potential military applications.

Additionally, Anthropic detailed five case studies involving individuals attempting to use Claude for biological weapons development, including research on modifying the chikungunya virus and highly pathogenic avian influenza. The report highlights the risks associated with AI model misuse and urges AI labs to pause development, warning that humans risk losing control over such technologies.