A weapons development cell in northern Yemen used Anthropic's Claude AI to write guidance software for a guided rocket and for a ballistic missile with a stated range of more than 2,000 kilometers, the U.S. artificial intelligence company said in a threat intelligence report published Thursday.
Anthropic said it had no evidence the group succeeded in fielding an operational weapon. The company banned the accounts involved and shared threat information with government and industry partners.
The report does not name the group. Northern Yemen is controlled by the Iran-backed Houthis, who have repeatedly launched ballistic missiles and drones at Israel since the start of the war in Gaza.
According to Anthropic, the cell was working on three weapons programs at the same time. The first was a guided rocket built around a phone-class flight computer with terminal homing guidance. The second was a multi-stage ballistic missile with a stated range of more than 2,000 kilometers. The third was a missile family the group called the "R2000" set, which included a hypersonic glide vehicle variant.
The operators used Claude Code "in place of human software engineers," the report said, to develop the guidance, navigation and control software that keeps a flying vehicle stable and on course. They ran several instances of Claude at once and assigned each a separate role, with one writing code, another conducting research and a third reviewing the work. Their tasks included writing control and position-estimation software, tuning flight controls, running firmware builds and carrying out flight simulations.
The group eventually conducted a live test of the guided rocket in Yemen. The test appeared to fail, and within hours the operators returned to Claude to investigate what had gone wrong, Anthropic said.
The report said the operators repeatedly tried to get around Anthropic's safeguards by concealing the purpose of their work and splitting it across multiple sessions, so that no single conversation revealed the scope of the weapons program. Anthropic acknowledged that its safeguards blocked many of the group's requests but not all of them.
Some of the group's capabilities are likely to outlast the ban. Anthropic said the cell had already built an offline simulation toolkit that runs without Claude or commercial engineering software such as MATLAB.
The Yemen case was one of several in the report involving attempts to use Claude for conventional weapons development, alongside cases in China and Russia. The report covers activity Anthropic disrupted between December 2025 and August 2026.
The same report details three Iranian state-aligned influence operations that used Claude to plan what the actors described as "soft war" or cognitive warfare campaigns aimed at audiences inside Iran and abroad. Anthropic tied the operations to the Islamic Culture and Communications Organization, which operates under Iran's Ministry of Culture and Islamic Guidance; the Islamic Propaganda Office of Khorasan Razavi, which the report said ran a cognitive warfare command room out of a seminary in Mashhad; and the Bina Cultural Observatory of the Islamic Propaganda Organization.
According to the report, the actors used Claude to produce campaign plans, doctrine manuals, persona systems and target databases, and designed content so that state messaging would appear to come from independent voices. The actor linked to the Islamic Culture and Communications Organization described the role of Iran's cultural attachés as "not to be the narrator, but the director."
Anthropic linked one operation to a director-level official at the Bina Cultural Observatory, who used Claude to generate messaging in the voice of a spokesperson for Iran's Islamic Revolutionary Guard Corps. During a campaign around the 2026 war between the United States, Israel and Iran, that network falsely attributed claims to the Western think tanks CSIS, Brookings and RAND to make its messaging appear more credible.
The operations also produced content targeting the Bahá'í, a persecuted religious minority in Iran, and built databases naming international officials and Iranian opposition figures. Because Claude is blocked inside Iran, the actors used VPNs and foreign phone numbers to register their accounts, but Anthropic said they repeatedly disclosed their locations, institutions and roles in their conversations. The company said it removed all three accounts.
