A buried court filing just confirmed AI’s most uncomfortable chapter has already begun
The disclosure was buried inside a lawsuit about pollution in Black communities. What it revealed was something far larger than the case it came from.
A sworn statement filed by Cameron Stanley, the Pentagon’s chief digital and artificial intelligence officer, confirms that xAI’s Grok was used in US military operations against Iran. Those operations involved more than 2,000 munitions fired at 2,000 separate targets within a 96-hour window.
What the Pentagon filing actually said
Stanley’s statement was submitted as part of a lawsuit alleging that xAI’s data centers are illegally polluting communities of color. His declaration was intended to argue against disrupting Grok’s operations, but in making that argument it placed on public record details that had not been previously confirmed.
Stanley described Grok as one of only four AI models currently capable of supporting national security applications and one of just three equipped for critical operational work in top-secret environments. He framed the continued availability of Grok as a matter of paramount national importance. This appears to be the first time a senior US administration official has explicitly confirmed that Musk’s AI was used in active military operations against Iran.
Why the Grok disclosure shifts the entire conversation
The 2,000 munitions figure is significant on its own. But what the filing signals about AI’s current position inside US military architecture is the larger story.
The debate over AI in warfare escalated earlier in 2026 after reports emerged that the US military had used AI systems during an operation targeting Venezuelan President Nicolás Maduro. That episode triggered a public dispute with Anthropic, the AI company whose terms of service prohibit the use of its models for violent purposes, weapons development or surveillance. Relations between Anthropic, the Pentagon and the Trump administration have reportedly worsened since.
The tensions are not limited to AI companies. In April, more than 600 Google employees signed a letter to CEO Sundar Pichai urging the company not to allow the Department of Defense to deploy its AI in classified work. Workers cited concerns about systems capable of lethal decisions, surveillance applications and the potential for errors that cannot be reversed. It was not the first time Google employees had taken that position. In 2018, internal pressure led the company to withdraw from Project Maven, a Pentagon contract to apply AI to military drone footage analysis.
Those concerns now have a specific, named example attached to them.
The competitive context around the disclosure
Grok’s role in US national security operations arrives at a difficult moment for xAI. Yann LeCun, considered one of the foundational figures in modern AI research, recently described xAI as a failure and questioned whether it could compete with OpenAI and Anthropic over time.
The Pentagon’s filing suggests that competition in this space is being judged by a different standard than consumer benchmarks or product reviews. Whether a model can function in classified environments, process targeting data at speed and hold up under conditions where reliability has direct operational consequences is a measure that has nothing to do with writing emails or generating images.
What comes next for AI and for warfare
AI companies built their products for productivity and framed them for everyday life. The military has been evaluating those same systems against a different set of requirements.
The question that researchers and ethics advocates have raised for years, whether these tools would eventually support or make lethal decisions, is no longer abstract. The answer appeared in a court filing about a pollution lawsuit, and the number attached to it is 2,000.