In what appears to be the first explicit government admission of its kind, a senior Pentagon official has confirmed that Donald Trump's administration used Elon Musk's Grok AI chatbot to launch more than 2,000 munitions at approximately 2,000 distinct targets in Iran within a 96-hour window. Cameron Stanley, the Department of Defense's chief digital and artificial intelligence officer, made the disclosure in a sworn declaration filed as part of xAI's defense against an NAACP lawsuit alleging illegal air pollution from the company's Black community-adjacent data centers.

Grok Goes to War

Stanley described Grok as one of just four AI models "currently capable of supporting national security applications" and one of three products "equipped to support mission-critical operations" in classified environments. The Pentagon's reliance on xAI's technology runs so deep that the Department of Justice has warned a federal court that any ruling halting xAI's deployment, refinement, or upgrades would "severely" impact national security. The filing specifically references the Grok Gov Model—a suite of products engineered for federal use that Stanley claims contains features "found in no other frontier AI model."

Targeting with Maven

The targeting data feeding these strikes came from the National Geospatial-Intelligence Agency's Maven Smart System, which uses AI to aggregate intelligence on a dashboard for military decision-makers. While Maven doesn't explicitly create targets, analysts suggest it played a role in identifying potential strike points during Operation Epic Fury. Human error reportedly compounded the problem: officials failed to verify whether target maps were current before authorizing strikes. U.S. military investigators now believe American forces were likely responsible for an airstrike on a girls' school in Minab that killed at least 175 people, predominantly children—the deadliest civilian casualty incident since U.S. and Israeli operations began in February.

Legal Crossfire

The Pentagon's admission comes as Congress moves to impose guardrails on military AI deployment. Senator Kirsten Gillibrand has introduced legislation requiring human commanders to retain authority over life-and-death decisions and banning AI use for nuclear weapons, domestic surveillance, and autonomous weapon systems. "The most critical decisions affecting our national security and the lives of our service members must always be made by human beings, not unaccountable machines," she stated earlier this month. Meanwhile, xAI faces a separate legal battle over its Colossus 2 data center, where at least 57 gas-burning turbines allegedly operate without Clean Air Act permits—turbines that Stanley's filing describes as providing "critical surge" capacity for wartime scenarios.

Anthropic Draws the Line

Not every AI company has rolled out the red carpet for Pentagon contracts. Anthropic rejected a proposed deal after concluding the administration wouldn't guarantee Claude wouldn't be used for domestic surveillance or autonomous drones. The retaliation was swift: the Pentagon designated Anthropic a "supply-chain risk to national security," threatening the company's government contracts in what has become an ongoing legal dispute. This stands in stark contrast to xAI's privileged position—and raises questions about which AI firms are willing to compromise ethical boundaries for defense dollars.

Key Takeaways

  • The Pentagon officially confirmed using Grok Gov Model to fire 2,000+ munitions at Iran in a single 96-hour operation
  • Targeting relied on the NGA's Maven Smart System alongside human verification failures that contributed to mass civilian casualties
  • Senator Gillibrand's bill would mandate human control over lethal decisions and ban AI for nuclear/surveillance/autonomous weapons
  • Anthropic's refusal to guarantee Claude wouldn't be used for domestic surveillance or autonomous drones resulted in a national security designation

The Bottom Line

This isn't just about one administration using one chatbot—this is about the normalization of AI-assisted killing at scale, with zero meaningful congressional oversight. When xAI gets environmental waivers and Anthropic gets blacklisted for having ethical red lines, you've got a defense procurement system that rewards exactly the wrong behavior.