Pentagon AI chief says Grok was used to direct 2,000 strikes on Iran
The Pentagon’s chief digital and artificial intelligence officer, Cameron Stanley, said in a sworn federal court filing that a government version of Elon Musk’s Grok model was used to help direct US strikes during the Iran war, The Hill reported . According to the filing, the “Grok Gov Model” enabled US forces to deploy more than 2,000 munitions against 2,000 distinct targets within 96 hours during an operation the document calls Epic Fury, which Stanley defended as “a matter of paramount national security.”
The disclosure appears to be the first explicit acknowledgment by an administration official that a commercial chatbot was folded into the targeting chain for the strikes, which killed hundreds of people. The mechanism matters: large language models are prone to fabrication and are not built to be reliable under the evidentiary standards that target selection normally requires, so using one to compress thousands of targeting decisions into days raises direct questions about how much human review each strike received.
The filing has already drawn a legislative response. Sen. Kirsten Gillibrand (D-N.Y.) has introduced a bill that would bar the use of large language models without human oversight in decisions involving the use of force, detention or other “high-consequence actions” — a sign that the dispute over autonomy in lethal systems is moving from policy debate to concrete law.