πΎπͺπ€ Yemeni weapons cell used Anthropic's Claude to build missile software β report A weapons engineering cell in Yemen used Anthropic's Claude models to create guidance and control software for rockets and missiles, a new report from Anthropic titled "Detecting and countering misuse of AI: September 2026" shows. π The cell was juggling three weapons programs at once: a guided rocket with final-phase homing, a multi-stage ballistic missile aiming for a range beyond 2,000 km, and something called the "R2000" β which included a hypersonic glide vehicle variant. Instead of hiring a team of software engineers, they leaned on Claude Code as their entire development layer. They ran several instances of Claude side by side, each with a different job β one writing code, one doing technical research, one checking the others' work. β‘οΈ Anthropic's filters blocked plenty of their direct requests. So the operators got creative: they hid what the work was actually for, split their prompts across isolated sessions so no single log showed the full picture, and used other workarounds to stay under the radar. Anthropic says there is no evidence the group succeeded in fielding a working long-range weapon. That said, the cell did test-fire a guided rocket β and the test appears to have failed. Within hours, they were back in Claude running a failure analysis on the telemetry. Once Anthropic caught on, it banned the accounts and shared what it found with security partners in both the public and private sectors. But by then, the group had already compiled some offline simulation tools β meaning they could keep working without Claude. π Substack | Chat | @geopolitics_prime