As cited
Copy frozen at (site build).
research
Using AI for Weapons Development
Anthropic disclosed that threat actors in northern Yemen used Claude to develop guidance, navigation, and control software for multiple weapons programs, including guided rockets, ballistic missiles, and hypersonic glide vehicles. The actors employed multiple Claude instances with specialized roles, split work across sessions, and used evasion tactics to circumvent safeguards, though many requests were blocked. While the actors test-fired a guided rocket that failed, there is no evidence of successfully fielded operational devices.
Why it matters: Security teams and policymakers must recognize that large language models can accelerate weapons development capabilities and plan detection and response strategies accordingly; artificial intelligence (AI) (artificial intelligence) providers need to refine safeguards against distributed, multi-session attack patterns.
- Source published
- First seen by Cybersecurity Tracker