Houthi rebels have been exposed for using Claude Code instead of engineers in missile development; Anthropic has suspended the relevant accounts.
Anthropic released a threat intelligence report on Thursday, September 10 (Eastern Time), stating that the company discovered a “threat actor group” located in northern Yemen using Claude to conduct missile and guided rocket development.
Anthropic did not directly name the Houthi rebels, but since the Houthis control most of northern Yemen, some media outlets have linked the individuals in question to them.
More noteworthy is that Anthropic claims these individuals used Claude Code to perform tasks originally done by human software engineers and conducted a guided rocket test launch.
Claude Code was used to replace software engineers and analyze the reasons for the failure of guided rocket test launches.
Anthropic stated that the team has advanced three weapons development projects, including multi-stage ballistic missiles, multiple missile types, and guided rockets, and is using Claude to develop guidance, navigation, and control software for aircraft.
Anthropic wrote in its report:
“These operators use Claude Code instead of human software engineers to develop software for guidance, navigation, and control (GNC) to maneuver and stabilize the aircraft.”
Anthropic stated that they also used Claude to integrate the open-source autopilot system into a "phone-level" flight computer, writing control and position estimation software, adjusting control parameters, running firmware build processes, and conducting flight simulations.
However, AI did not help it successfully deploy the weapons. Anthropic stated:
“We have no evidence that these operators successfully deployed a usable weapon; however, they did test-fire a guided rocket.”
The company stated that the test launch "appeared to have failed," and several hours later, personnel returned to Claude to attempt to analyze the reasons for the failure.
The security measures blocked "many requests," but not all.
Anthropic stated that the individuals involved employed various methods to bypass Claude's security mechanisms, including concealing the true objective and end use of the software, and splitting the work into multiple sessions so that a single conversation could not reveal the full intent.
Anthropic stated:
"Our security measures blocked many of their requests, but not all of them."
The company subsequently banned accounts associated with these actors and shared threat intelligence with public and private sector partners. However, Anthropic also discovered that these individuals had previously developed an offline simulation tool that did not rely on Claude or other engineering computing environments .
“Even so, we have evidence that these operators have also developed an offline simulation tool that does not rely on Claude or other engineering computing environments.”
It must be emphasized that Anthropic did not directly identify these operators as the Houthi rebels, nor did it provide evidence that the AI-assisted research and development ultimately resulted in a missile system that could be deployed in actual combat .
What has been confirmed so far is that Anthropic has discovered a weapons development group in northern Yemen that closely matches its description, and confirmed that the group used Claude to develop guided weapon software, conduct guided rocket test launches, and perform AI-assisted analysis after test launch failures.
Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.