Grok used to direct U.S. strikes against Iran, Pentagon confirms under oath

The disclosure, made in a legal briefing, marks one of the most explicit public admissions of a commercial LLM being used in lethal military targeting. A Pentagon AI chief testified that Grok is integrated into Project Maven, the U.S. military's AI-assisted targeting program, and that it facilitated significant operational efficiency during 'Operation Epic Fury' against Iran. One official went further, claiming Grok helped fire 2,000 missiles — a figure that has not been independently verified.

The revelation surfaced amid an environmental lawsuit targeting the natural-gas turbines powering xAI's data centers. In an unusual move, the DOJ intervened on xAI's side, citing Grok's military role as a national-security justification for keeping the turbines running. That coupling of commercial AI infrastructure with defense priorities is itself a notable escalation in how the government frames AI compute as strategic.

The story compounds xAI's mounting governance problems. A separate lawsuit alleges the company fired engineer Devin Kim after he raised safety concerns about Grok. Developers and ethicists reacted with alarm to the idea of an LLM — particularly one as loosely guardrailed as Grok — being placed in a targeting loop. The episode lands in the same week as the Anthropic shutdown, sharpening a week-long theme of AI and the national-security state: one model pulled for being too dangerous, another celebrated for its battlefield utility. Watch for congressional questions on oversight of autonomous-targeting systems and whether xAI's competitors face pressure to disclose their own defense entanglements.