Anthropic, the developer of the Claude chatbot, disclosed that accounts it traced to the Houthi‑controlled highlands of northern Yemen attempted to employ its AI system to create guidance software for advanced missile projects. The company blocked the accounts after identifying the activity.

AI misuse discovered

According to Anthropic’s September report, the users conducted a test of a guided rocket that did not succeed. The company learned of the failure because the same accounts returned to Claude for troubleshooting advice. While the test did not result in an operational device, the report notes that the actors were pursuing three separate weapons programs, one of which involved a multi‑variant missile capable of hypersonic glide.

Houthi claims and capabilities

Hazam al‑Assad, a member of the Houthis’ political bureau, described the effort as “unreasonable and illogical,” insisting that the rebel forces already possess modern, diverse production capabilities acquired during years of conflict with Saudi Arabia. He framed all weapons use as self‑defense.

U.S. weapons analyst Trevor Ball of Armament Research Services cautioned that, despite the Houthi interest in hypersonic technology, the group lacks the industrial base to manufacture such systems. Ball noted that even the United States is still testing its own hypersonic missiles.

Iranian support and regional impact

Iran has publicly claimed possession of hypersonic missiles and alleged use against Israel, though it denies directly arming the Houthis—a violation of a United Nations embargo. Iranian‑made anti‑ship missiles with mid‑course guidance have already been documented in Houthi arsenals.

The Houthis have intensified attacks on Saudi oil facilities and Red Sea tankers, threatening global trade routes. Recent territorial gains along the Red Sea coast and the Bab al‑Mandeb Strait have heightened concerns about the group’s expanding reach.

Broader AI‑enabled threats

This is Anthropic’s third report since March 2025 on the misuse of its AI platforms. Earlier disclosures covered state‑sponsored propaganda campaigns and attempts to enhance biological weapon designs. In the Yemen case, the actors reportedly built an offline simulation toolkit that did not rely on Claude or other cloud services.

Anthropic says it has shared its findings with both private and public partners to help mitigate the risk of AI being weaponized by hostile actors.

Original reporting: Alexandria, VA News – WTOP News — read the source article.