Anthropic says a military group in northern Yemen attempted to use its Claude AI model for missile-related software and engineering tasks.
A military group in northern Yemen attempted to use artificial intelligence to assist with missile development, according to a new intelligence threat report by U.S. AI company Anthropic.
The company did not directly identify the group as Yemen’s Iran-backed Houthis. However, its report described an armed group operating in northern Yemen that was running three separate weapons-development programs. Most of northern Yemen is controlled by the Houthis.
According to Anthropic, the group was working on multistage ballistic missiles, other types of missiles and guided rockets. Rather than relying solely on conventional software engineers, the group attempted to use Anthropic’s Claude AI model to help develop software related to guidance, control and navigation.
Media reports citing the Anthropic assessment said AI was used in efforts to connect an open-source autopilot to a phone-class flight computer, improve control settings and conduct flight simulations.
Anthropic said its safety systems blocked multiple requests made by individuals associated with the group. However, the individuals reportedly used different methods to conceal their broader objectives.
The company said the activities were divided across multiple sessions, apparently making it more difficult for a single conversation to reveal the full purpose of the requests.
Anthropic said there was no evidence that the group had successfully developed fully operational weapons. However, the assessment indicated that a guided rocket was tested.
The rocket test apparently failed. Several hours later, the individuals reportedly returned to Claude and sought assistance in determining why the test had failed.
Anthropic subsequently banned the accounts involved and shared the information it obtained with relevant government and private-sector organizations, according to the report.
The assessment also highlighted broader concerns about the potential misuse of AI technology. Anthropic said some actors had attempted to use Claude for research related to biological weapons.
Another concern was the reported development of an offline simulation toolkit by the group. According to media reports, the toolkit was designed to operate without relying on Claude or other engineering platforms.
About The Author
