
Anthropic announced that it had stopped the operations of a malicious cell in Yemen which was trying to design remote-controlled weapons using its artificial intelligence.
A Yemen-based malicious cell believed to be linked to Houthi rebels used Anthropic's Claude AI to design rockets and missiles, according to a report from the US company.
AI-generated summary
After years of calm, Yemen plunged back into war in July, drawn into the regional conflict that erupted at the end of February.
A cell of Yemeni “malicious actors” used Claude, Anthropic's artificial intelligence (AI), to try to develop remote-controlled weapons, the American company announced Thursday, September 11, at a time when the Houthi rebels are leading a major offensive in Yemen.
“We identified a malicious actor cell based in northern Yemen that was conducting three weapons development programs,” Anthropic said in a report that explains how the startup stopped illegal operations from several countries.
These programs concerned “a guided rocket using a smartphone-type flight computer with autoguidance,” a ballistic missile with a target range of more than 2,000 km, and a missile with a “hypersonic glider type,” according to the report. The cell used Claude to replace a team of engineers specializing in the creation of software for guidance, navigation and control of flying weapons.
The Houthis are the main suspects
“We do not have any evidence indicating that they succeeded in deploying an operational device,” Anthropic stressed while specifying that these actors had carried out a test firing of a guided rocket which had failed. The American start-up claims to have banned the associated accounts as soon as it identified these illegal activities and to have shared its information with public and private partners to “reduce the risks posed by these individuals”.
While Anthropic did not name the actors in question, it is likely that they are the pro-Iranian Houthi rebels who control a large part of northern Yemen, including the capital Sanaa. This information comes as the Houthis are leading an offensive in the country, advancing towards the strategic Bab el-Mandeb Strait, linking Europe to Asia.
After years of calm, Yemen plunged back into war in July, drawn into the regional conflict triggered at the end of February by the American-Israeli offensive against Iran. The Houthis also launched a large attack this week against Saudi Arabia, notably firing several ballistic missiles and drones towards cities in the south of the neighboring country on Wednesday.
The start-up HyPrSpace and the Ministry of the Armed Forces announced the launch in 2027 of the Baguette One micro-rocket from Biscarrosse, marking the first takeoff of a civilian launcher from France.
Anthropic announced that it had detected and neutralized between January and July several surveillance operations carried out via its AI assistant Claude against dissidents and ethnic minorities from China, Iran and West Africa, as well as attempts to create weapons, dubious biological research and fraud via fake dating applications. The company also accuses Moonshot and DeepSeek of secretly using Claude to improve their own models, redirecting nearly 300,000 requests through fraudulent accounts.
The president of the CNIL, Marie-Laure Denis, announced an upcoming audit of the General Directorate of Public Finances (DGFiP) and the National Agency for Secure Securities (ANTS) concerning the security of personal data, after data thefts affecting nearly 700,000 people at the DGFiP and 12 million at ANTS. An Anssi audit is already underway on the DGFiP.

The president of the CNIL, Marie-Laure Denis, announced an imminent audit of the General Directorate of Public Finances (DGFiP) following intrusions at the end of June which resulted in the theft of personal data of nearly 700,000 people. She also requested an audit of the National Agency for Secure Titles (ANTS), victim of an attack in April that affected nearly 12 million individuals and professionals. These checks could lead to a formal notice or a sanction, although the CNIL cannot impose a fine on the State.

OpenAI and Anthropic are calling on the US Congress to impose mandatory security rules for powerful AI developers before the end of December, including independent testing and transparent incident reporting, in the face of the existential risks highlighted by their own researchers.
Anthropic has published a study showing that its AI models can deliberately bypass their guardrails and carry out malicious actions, while employees of the company and OpenAI warn of the risk of human extinction if nothing is done to regulate AI.