The artificial intelligence company Anthropic reported in a threat intelligence brief published this week that its Claude AI model was used by a group in Yemen for research related to weapons development. According to the company, actors associated with a weapons development cell in northern Yemen used the AI to inquire about the design of guidance systems for missiles and rockets before their access was terminated.
The findings were detailed in Anthropic’s report, “Mapping the Misuse of AI,” which outlined the company’s efforts to identify and stop malicious state-aligned actors from using its technology. The report stated that the Yemen-based group leveraged the Claude model for tasks that included requesting information on satellite communication components and seeking technical assistance with software code relevant to unmanned systems.
Anthropic characterized the activity as “low-sophistication” and noted that its safety models were effective in preventing the AI from providing sensitive or dangerous information. The company reported that it promptly detected the activity and shut down the associated accounts.
“In all of the cases described in this report, our safety systems were successful in detecting the violative content and shutting down the identified accounts,” the Anthropic report stated. The company added that the actors’ use of their model “did not appear to have yielded any meaningful results.”
The brief also documented similar attempts at misuse by other state-linked groups. According to Anthropic, it disrupted activity from a North Korean group known as APT43, which used the AI platform to research publicly known vulnerabilities, draft phishing emails, and seek information on sanctions. A separate set of actors linked to the Russian government reportedly used the models to generate content for information campaigns.
The disclosure comes as technology companies and government agencies express concerns over the potential for advanced AI models to be exploited for malicious purposes, ranging from cyberattacks and propaganda to weapons proliferation. Major AI developers, including Anthropic, OpenAI, and Google, have established safety policies and monitoring systems intended to prevent such activities on their platforms.
In its report, Anthropic stated that publishing its findings contributes to transparency and helps the broader security community develop countermeasures. The company confirmed it would continue monitoring its platforms for misuse and collaborating with industry and government partners to address emerging threats.
