Cameron Stanley's sworn declaration calls Elon Musk's xAI model 'paramount national security,' while the same campaign is tied to a girls' school strike that killed more than 150 children.
Cameron Stanley, the Pentagon's chief digital and artificial intelligence officer, has put a number on Elon Musk's xAI chatbot Grok for the first time in a public filing. In a sworn declaration in NAACP v. xAI, Stanley said the Pentagon used Grok to fire 2,000 munitions at 2,000 distinct targets within 96 hours during the Iran campaign, calling the system "a matter of paramount national security" (Stanley sworn declaration, NAACP v. xAI, S.D. Miss., No. 3:25-cv-52261, doc 58).
The figure, first reported by The Independent and Futurism, comes from the officer whose job is to evaluate AI for the entire Defense Department. It is not a Pentagon press release. It is a court record, sworn under penalty of perjury, in a case about whether xAI's Colossus 2 data center in Mississippi is illegally polluting the air around a historic Black community.
Stanley framed Grok as one of four AI models "currently capable of supporting national security applications" and one of three "equipped to support mission-critical operations" in top secret settings, with the Pentagon relying on xAI's Grok Gov Model suite (Stanley declaration). The same declaration describes target identification running through the National Geospatial-Intelligence Agency's Maven Smart System, which surfaces points of interest for human commanders rather than auto-selecting targets. That posture, humans in the loop and AI surfacing options, is the framing the Pentagon is asking a federal judge in Mississippi to credit as the Justice Department and Defense Department press to dismiss the NAACP's Clean Air Act suit against xAI.
The same Iran campaign has produced one of the deadliest single-strike civilian casualty events publicly tied to the US-Israeli operation. A March 2026 strike on the Shajareh Tayyebeh girls' school killed at least 156 people, mostly children, according to Amnesty International, which is investigating the incident as a possible unlawful attack. The Independent reported a higher toll of 175 from the same strike. Stanley's filing does not link Grok directly to that strike. The court record places Grok inside the same campaign that includes it.
The political backdrop is harder to read. Anthropic's Claude was previously reported to have been used in the Iran strikes. After negotiations over Claude's role in domestic surveillance and autonomous weapons collapsed, the Pentagon designated Anthropic a "supply-chain risk to national security," a designation Anthropic is challenging in court. Sen. Kirsten Gillibrand's pending bill would require a human commander in the loop for life-and-death decisions and bar AI from nuclear, domestic-surveillance, and autonomous-weapons roles. Both the designation and the bill remain proposed or contested, not enacted policy.
For now, the public record on Grok's role in the Iran campaign rests on one man's signature. Stanley's declaration gives the operation its first concrete AI benchmark, 2,000 targets in 96 hours. The same document asks a federal court to treat Grok as decision-support inside Maven, not as the decision-maker. Whether that distinction holds up under the next civilian-casualty investigation, or under the next round of the Mississippi litigation, is the open question.