US military admits using Musk’s AI to target Iran

Dado Ruvic / REUTERS

Grok, Elon Musk’s AI, helped the US army in its war in Iran (illustrative image).

Artificial intelligence has become a weapon of war. The US government revealed that Elon Musk’s artificial intelligence, Grok, was used in the strikes against Iran, in a legal brief that defended the multi-billionaire’s company, xAI, targeted by an environmental complaint over gas turbines in a data center giant.

The Ministry of Justice, in a brief filed Sunday June 15, argues that this complaint “Menace to national security, economy and energy” of the country, risking cutting off power to AI infrastructures now used by the army.

To support this argument, the department called Cameron Stanley, head of AI at the Pentagon, to testify. He declares under oath that a tool derived from Grok, the « Grok Gov Model »is already deployed within Project Maven, the army’s AI-assisted targeting program, initially based on Anthropic’s Claude model.

Plus de 2 000 munitions sur 2 000 cibles

According to this statement, Maven processes “Enabled US forces to deploy more than 2,000 munitions on 2,000 separate targets in 96 hours.”during the war against Iran. The senior official sees in these figures “testimony of a very large increase in operational efficiency made possible by the Grok Gov Modelâ€without specifying if it is the only model used by this program.

Stanley says Maven users consume “About 2 billion tokens” (calculation unit) “par jour”either « jusqu’à 6 millions de pages » processed, a volume which, according to him, makes xAI’s calculation infrastructure essential.

The turbines targeted by the complaint power Colossus 2, an xAI supercomputer which drives Grok, on the outskirts of Memphis. The NAACP, an association defending black rights, is suing xAI, accusing it of operating dozens of turbines without a permit, in violation of the Clean Air Act. They pollute predominantly black neighborhoods, says the association. xAI maintains that its turbines are temporary and mobile, and therefore are not subject to these regulations.

At the end of February, the government broke its contracts with Anthropic, which refused to allow its tools to be used for fully automated strikes or mass surveillance of the Americans. The Pentagon then turned to Google, OpenAI and xAI, the three other American cutting-edge AI companies. But this transition takes time and the government had to admit in March that Claude continued to be used for the war in Iran.

The military use of AI is controversial. At Google, more than 600 employees demanded in April not to provide AI to the army for classified operations. The group had already given up on Maven in 2018 under pressure from its engineers.