Anthropic has reported significant interventions involving its Claude artificial intelligence model, which has thwarted attempts to create software for missiles and guided rockets. Between December 2025 and August 2026, the San Francisco-based AI lab identified and disrupted six operations aimed at developing conventional weapons and conducting intelligence gathering. These efforts were linked to actors in Yemen, Russia, and China.
Anthropic said it has disrupted six operations for the development of conventional weapons in Yemen, Russia, and China since December.
Anthropic confirmed that the actors did test one of the guided rockets but said it had no evidence that they succeeded. “We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket. The three tracks involved drafting Chinese-language specifications for the torpedo system, producing technical proposals, and benching their own system against specific US anti-torpedo and anti-submarine programs – all using Claude. Anthropic pointed out a Russia-based operation that used Claude to engineer an autonomous military drone swarm. It also claimed to have disrupted another Russian operation to procure mixed military and civilian goods using Claude.
In its report about the misuse of its models, Anthropic identified what it called a “Yemen-based weapons engineering cell” behind one such attempt to develop the software that could control a flying vehicle. The threat actors in Yemen had three weapons development programs, including a multi-stage ballistic missile with a range above 2,000 km and a missile system referred to as the “R2000″ set, which included a hypersonic glide vehicle variant. Read: Scientists Tried To Use AI Models For Potential Biological Weapons Research: Anthropic They used Claude Code instead of human engineers to build software that could guide, navigate, and stabilise a flying vehicle, the AI lab said. This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed,” the company said. Anthropic said it banned the accounts associated with such actors and shared threat information with its partners to mitigate the potential risks. In another case, it said it had disrupted a China-based operation that used Claude to draft a fire control specification and acquisition documents for undersea warfare. “We identified a China-based threat actor who used Claude to advance three parallel tracks of work on an anti-torpedo weapons system,” it said. Read: OpenAI Seeks US Rules On Powerful AI Amid Anthropic Researcher’s Warning Two other Chinese operations – which used Claude to build targeting software for electronic warfare and collect intelligence on directed-energy weapons and their supply chain – were also disrupted, the company said. The actors used Claude Code to write and test the code to govern attack, observe, and return-to-base behaviours, the company said.
