Report findings: Anthropic’s September 10, 2026 threat-intelligence report examined malicious uses of Claude detected between December 2025 and August 2026. The 154-page report covers weapons development, surveillance, propaganda and cyber operations. Anthropic said the cases were not typical examples, but were among the most notable and novel forms of misuse it had identified.
\n\nWider reach: The report examined seven areas of harm and detailed six weapons-related cases. It also identified nine influence operations targeting audiences across six continents. Anthropic said AI can reduce the expertise, manpower and time needed for sophisticated operations. The cases showed smaller groups using AI for tasks that previously required teams of specialists.
\n\nYemen weapons: In northern Yemen, Anthropic identified a cell that used Claude across three weapons programmes. Claude Code was used instead of human software engineers to help develop software that steered and stabilised a weapon. The programmes included one guided rocket, one ballistic missile with a stated range above 2,000 km, and a missile family containing a hypersonic-glide variant.
\n\nAI as engineers: The Yemen group used several Claude instances for different engineering tasks. One instance wrote code, another researched information and a third checked the code. The group later test-fired a guided rocket.
Original source: https://www.cnbctv18.com/technology/
