Elon Musk’s Grok artificial intelligence platform was used to execute targeting operations against Iran, the Pentagon’s top AI official confirmed under oath — a disclosure that surfaced not through a press briefing or congressional hearing but buried inside a federal court filing defending an xAI data center’s right to run unpermitted gas turbines in a Memphis neighborhood.
The sworn testimony, submitted June 15 as part of a Justice Department brief supporting xAI in an environmental lawsuit brought by the NAACP, places Grok inside Project Maven, the military’s AI-assisted targeting program that has become central to U.S. combat operations. Cameron Stanley, the Pentagon’s chief AI officer, stated in his declaration that Maven Smart Systems — the operational layer through which AI targeting runs — enabled American forces to deploy more than 2,000 munitions against 2,000 distinct targets within a 96-hour window during Operation Epic Fury.
That is not a marginal support role.
Stanley’s declaration praised the increased operational efficiency the Grok Gov Model brought to the campaign — language that, in a different context, might have appeared in a Pentagon press release.
Read also: Elon Musk Found Guilty Of Misleading Twitter Shareholders
Instead it was written for a judge weighing a civil pollution case, and in making the national security argument explicit, federal prosecutors placed operational details about the Iran campaign into the public record through a civilian docket. The Justice Department’s brief framed the stakes plainly: the lawsuit, it argued, threatens American national, economic and energy security by seeking to cut the power supply for AI innovation that the Defense Department relies on for its military operations.
The relationship between Grok and the Iran campaign did not enter the public record on the military’s own terms. It was a byproduct of litigation. The NAACP sued xAI alleging that dozens of gas turbines powering its Memphis data center were operating without the required permits in violation of the Clean Air Act, disproportionately burdening majority-Black communities surrounding the facility. xAI has maintained the turbines are temporary and mobile and therefore exempt from regulation. The government’s decision to invoke national security in defending that position — drawing a direct line from Memphis to Tehran — turned a neighborhood pollution dispute into an inadvertent window onto classified military operations.
The path from the Iran campaign to a Tennessee federal court runs through several months of upheaval in the military’s AI vendor landscape. Grok’s predecessor in Project Maven was Anthropic’s Claude model, which held the primary contract until late February, when the government severed ties after Anthropic declined to permit its technology to be used for fully automated strikes or the mass surveillance of American citizens. The refusal was a line the company drew explicitly, and the Pentagon moved quickly in response — turning to Google, OpenAI and xAI to fill the gap. The transition was not seamless. By March, the government acknowledged that Claude remained in use during the Iran campaign even after the contract had been cut.
The expanding battlefield role for AI has not gone unchallenged inside the companies being recruited to fill it. More than 600 Google employees signed a petition demanding the company refuse to provide AI for classified military operations, one expression of a broader unease that has surfaced at several major tech firms as the scope of automated targeting has widened.
Musk occupies a singular position in this arrangement. He merged xAI into SpaceX in February — the same month Anthropic’s Maven contract was terminated — and SpaceX completed what was reported as the largest initial public offering in U.S. history on June 12. His relationship with President Donald Trump, whose administration is simultaneously prosecuting the Iran campaign and defending xAI’s turbines in federal court, has sharpened questions about the concentration of commercial and national security interests in a single set of hands.
None of that political architecture appeared in Stanley’s court declaration. His testimony was narrowly framed around operational performance — targets acquired, munitions deployed, efficiency gained.
The document was prepared for a civil proceeding, not a Senate committee, and it entered the public record because a civil rights organization sued over diesel exhaust. What it disclosed, as a matter of incidental record, was that within a four-day window during Operation Epic Fury, American forces armed with AI-assisted targeting struck 2,000 distinct points across Iran.
The NAACP’s lawsuit continues. So does the war.