An internal investigation by Anthropic, the artificial intelligence company behind the Claude chatbot, has uncovered that operatives linked to the Houthi movement in Yemen have used its AI systems to gather technical knowledge relevant to missile development. The finding, detailed in a newly published report, marks one of the most stark examples yet of how commercially available AI tools are being tapped by militant groups seeking weapons expertise.
According to the report, users associated with the group engaged Claude in conversations that sought information on propulsion, materials science, and engineering principles connected to rocket and missile design. Anthropic researchers emphasized that the AI models they deploy were not explicitly designed or intended to assist with military applications, but the openness of large language models means that such queries can be phrased in ways that bypass standard safety guardrails.
The revelation comes at a moment of heightened concern across the global AI industry about dual-use risks the growing accessibility of capable chatbots. Anthropic has repeatedly stated that it monitors for abusive patterns and works with external researchers and government bodies to understand how its products are being used. However, the report underscores a persistent challenge: even when companies implement content filters and usage restrictions, determined users can often rephrase requests to extract useful information.
The Houthi movement, officially known as Ansar Allah, has been engaged in a prolonged conflict in Yemen backed financially and militarily by Iran. Over the past several years, the group has demonstrated an expanding arsenal of drones and short-range missiles, much of it assembled through local workshops and imported components. Analysts say the use of civilian AI tools could accelerate the pace at which such groups acquire technical know-how that was previously accessible only through formal training or illicit networks.
Anthropic did not disclose specific dates or quantify exactly how many Houthi-linked accounts were involved. The company said the report was shared with relevant authorities and called for broader industry cooperation on monitoring and mitigating the misuse of AI in conflict zones.
