Russians built a drone swarm with Claude that picks its own targets.
Anthropic published its most detailed misuse report on Thursday, covering attacks it caught since December 2025. Russia-based freelancers used the coding tool Claude Code to build kamikaze drones that pick targets and detonate alone. Their targeting model trained on combat footage scraped from Ukraine, and Anthropic says it shut every operation down.
Why it mattersAI has collapsed the skill gap, so one person can now run a state-sized hacking campaign. The same agent tools you use for work are on sale to everyone else too.
Read in the app — free, in 9 languages
Related stories
Visa, Mastercard and Ant agreed on how to check an AI shopper.2026-09-11The robot brain that learns by watching hit $100m a year.
2026-09-11A Senate chairman called OpenAI's rogue-agent testing reckless.
2026-09-11GPT-6 Astra solved the last problem in maths' hardest test.
2026-09-11OpenAI opened the engine behind Codex to any developer.
2026-09-11