A sworn statement filed Monday in Mississippi federal court by Cameron Stanley, the Department of Defense's chief digital and artificial intelligence officer, says xAI's Grok Gov Model "enabled U.S. forces to deploy over 2,000 munitions to 2,000 distinct targets within 96 hours during Operation Epic Fury." That single sentence makes Grok the first frontier LLM officially named in a federal filing as the analytical layer behind a live strike campaign.

The disclosure is also, structurally, an accident. Stanley's affidavit isn't a Pentagon press release. It was submitted as evidence in the administration's bid to dismiss a NAACP Clean Air Act suit targeting xAI's Colossus 2 data center near Memphis, where the NAACP alleges at least 57 unpermitted gas turbines power the model. xAI says the turbines are temporary, mobile, and exempt from state air-permit rules for up to a year. To defend that posture in court, the DoD had to argue national-security necessity, and to argue necessity, it had to describe the use case. The combat record is, in effect, a procurement brag entered into the docket.

The procurement frame is the real story. Stanley describes Grok Gov as one of "just four proprietary AI models currently capable of supporting national security applications" and "one of just three enterprise providers equipped to sustain mission-critical operations across Secret and Top-Secret classified networks," with "features unique to xAI that are found in no other frontier AI model." Without it, he writes, "DoW's ability to meet its national security mission and keep pace with adversaries will be impaired."

Inside the National Geospatial-Intelligence Agency's Maven Smart System, the model handles targeting, intelligence, readiness, recruitment, planning workflows, report synthesis, predictive logistics, red-teaming, personnel management, and medical supply lines. Stanley's filing, as Yahoo notes, is careful: Grok and its peers "do not explicitly create targets" but feed Maven's dashboards with points of interest. CENTCOM has separately claimed a target generation rate of 1,000 per hour.

The body count complicates the brag. The Lieber Institute at West Point, cited by The Debrief, puts Iranian civilian deaths at roughly 1,700 as of May 2026. Among them: at least 175 people, mostly children, killed when Operation Epic Fury opened on Feb. 28 with a strike on the Shajarah Tayyebeh school in Minab, attributed by U.S. investigators to outdated targeting data routed through Anthropic's Claude. The DoD has since cut ties with Anthropic over its refusal to drop safeguards against lethal-autonomy and domestic-surveillance use, and designated the company a "supply chain risk to national security." xAI, acquired by SpaceX in February 2026, faced no such friction.

That sequencing is the disclosure's quiet center. A safety-pilled vendor gets blamed for a school and ejected from the stack; the vendor that absorbs the workload gets named, on the record, as indispensable.

Sources

https://www.yahoo.com/news/politics/articles/pentagon-used-elon-musk-grok-193053864.html
https://thedebrief.org/grok-goes-to-war-pentagon-reveals-musks-ai-chatbot-launched-missiles-in-u-s-war-with-iran/
https://gizmodo.com/using-grok-to-bomb-iran-and-the-twisted-dream-of-causing-death-without-killing-2000773399
https://thehill.com/policy/technology/5928204-pentagon-musk-grok-chatbot-iran-strikes/
https://www.washingtontimes.com/news/2026/jun/17/military-used-elon-musks-grok-strikes-iran-pentagon-reveals/