The Pentagon says that Grok helped hit 2,000 targets in 96 hours and that the polluting power plant that supports it is a matter of the highest national security. This admission came not in a press release or briefing from the Pentagon, but in a statement filed in federal court in Mississippi in an air pollution case. There, defending xAI against a Clean Air lawsuit, the Pentagon’s chief digital and artificial intelligence official recorded a sentence that should give anyone reading it pause.
See also: Former Swiss president files defamation lawsuit against Grok
The chatbot known as Grok had helped launch over 2,000 munitions on 2,000 different targets in Iran in 96 hours, and its continued operation was a matter of utmost national security. A government official revealed that a consumer AI product had been used to bomb a country, not to inform the public, but to keep a data center running.
The targeting and the turbines appeared in the same affidavit because, according to the administration’s narrative, they make the same argument: Grok is important to the war, the war is important to the nation, so the power plant that powers Grok cannot be shut down, no matter what the law says about its permits.
This is where the two problems with military AI cease to be separate. The first is that a chatbot built to answer questions on a social network is now connected to the machine that kills people. The second is that the human who owns that chatbot is also inside the government deciding how it will be used.
Each of these is troubling in its own right. Together they describe a system in which the safeguards we would normally expect—legal, environmental, ethical—are all bent toward the convenience of one company and one person.
Grok is, according to the filing, one of only four AI models the Pentagon considers capable of supporting national security applications and one of three approved for mission-critical operations in top-secret environments. It powers the Maven Smart System , the AI-powered dashboard that presents intelligence data to help officials decide what to strike.
See also: Ireland investigates X over Grok's "sexual images"
The official line is reassuring in its wording: AI doesn't create targets; it identifies points of interest for human analysts to evaluate. Humans, we are told, stay in the loop.
The circuit did not save Minab’s children. On February 28, in the first wave of attacks, a Tomahawk missile struck the Shajarah Tayyebeh girls’ school in the Iranian city, killing dozens of people, more than a hundred of them children by most estimates, in what observers have called the deadliest single incident of civilian casualties in the conflict.
A preliminary investigation by the US military concluded that US forces were likely responsible and that the attack was based on outdated information: target coordinates generated from old data that were passed through the system, never checked again with a current map. The school was a school all along.
That’s the problem with putting AI close to the target and having nothing to do with whether a machine pulls the trigger. The risk is more subtle and harder to legislate against. An AI dashboard that ranks points of interest and matches them to weapons doesn’t remove human judgment so much as it refines it. The analyst sees a confident interface, a clean readout, a target that the system has already rated with a number, and the natural human reaction is to trust it.
See also: Grok AI will curate your timeline on X
Automation bias is not a flaw in these systems; it is the predictable result of building something that is designed to look authentic.
