Anthropic Suspected of Houthi Rebels’ Attempt to Develop Guided Weapons Using Claude

[Financial News] Anthropic said it detected and blocked signs that an armed group in Yemen had attempted to use its conversational AI chatbot, Claude, to develop guided rockets and ballistic missiles.
According to a report on preventing the illegal use of its models released by Anthropic on the 10th (local time), a threat group based in northern Yemen was found to have used Claude without authorization while pursuing three weapons-development programs, AFP reported on the 11th.
The weapons they reportedly sought to develop included a terminally guided rocket equipped with a flight computer comparable to a commercially available smartphone; a multistage ballistic missile with a range exceeding 2,000 km; and a multirole missile incorporating a hypersonic glide vehicle (HGV) variant.
The group primarily used Claude to write guidance, navigation, and control (GNC) software, a task requiring a team of skilled engineers. Anthropic explained, "There is no evidence that the weapons were successfully deployed, but a test launch of the guided rocket was conducted." It added, "The test launch appears to have failed, and several hours after the launch, the group accessed Claude again to analyze the cause of the failure."
Anthropic permanently suspended the associated accounts immediately after detecting the illegal activity.
Although the report did not identify the specific actor, Iran-aligned Houthi rebels, who control northern Yemen and the capital, Sana'a, are widely considered the most likely group to have been involved. The Houthi rebels are currently mounting a full-scale offensive toward the Bab-el-Mandeb Strait, a strategic chokepoint linking Europe and Asia. This has heightened international concerns over attempts to develop weapons through the misuse of advanced technology.



[email protected] Yoon Jae-jun Reporter