The Vertical

One Man. One Stack. One War.

The Pentagon’s AI targeting chief disclosed in sworn federal testimony that Elon Musk’s Grok helped fire 2,000 munitions at 2,000 targets in 96 hours. The disclosure didn’t come in a press briefing. It came in an environmental lawsuit. That’s not a footnote. That’s the whole story.

I. The Admission Nobody Planned to Make

On June 15, 2026, the U.S. Department of Justice filed a brief in a federal courthouse in Mississippi. The case was about air pollution — specifically, whether xAI’s Colossus 2 data center in Memphis was illegally running dozens of gas-burning turbines without the permits required under the Clean Air Act. The NAACP sued. The DOJ intervened to get the case thrown out.

To make its argument, the Pentagon’s chief digital and artificial intelligence officer, Cameron Stanley, submitted a sworn declaration. In it, he disclosed that xAI’s Grok — a consumer AI chatbot — in its government-tuned form had been integrated into Project Maven, the military’s AI-assisted targeting system. And that during Operation Epic Fury, Maven Smart Systems “enabled U.S. forces to deploy over 2,000 munitions to 2,000 distinct targets within 96 hours.”

One target every three minutes. Around the clock. For four straight days.

The government disclosed that a consumer AI product had been used to bomb a country — not to inform the public, but to keep a data center from being shut down for Clean Air Act violations.

That last part is the one that should stop you cold. This wasn’t a transparency measure. It wasn’t a congressional hearing. It wasn’t a Pentagon press briefing. The American public learned that Musk’s AI helped orchestrate one of the most intensive air campaigns in recent history because the government needed a judge in Mississippi to let an unlicensed power plant keep running.

National security, in other words, has become the universal solvent. It dissolves environmental law. It dissolves the objections of families living downwind. It dissolves accountability — and it does all of this in service of a private company’s unpermitted infrastructure.

II. The Stack

To understand what’s actually happening here, you need to map the vertical.

Elon Musk owns SpaceX — the rockets and the Starlink satellite network that carries military communications. He merged xAI into SpaceX earlier this year, making Grok part of the same corporate entity. The Colossus 2 data center in Memphis powers Grok’s training and inference. Starlink satellites carry the data. Grok Gov processes the intelligence. Maven Smart Systems displays it on a targeting dashboard. A human commander approves the strike.

In theory, humans remain in the loop. In practice, the loop runs on Musk’s infrastructure at every step.

Stanley’s testimony confirmed: Grok is one of only four AI models the Pentagon considers capable of supporting national security applications. It is one of three cleared for mission-critical operations at the Secret and Top Secret classification levels. It handled threat identification, target prioritization, and predictive analysis — every stage of the kill chain before the final human authorization.

One man owns the rocket, the satellite, the AI model, the data center powering it, and the coding tools developers use to build on top of all of it. He is simultaneously the vendor and the infrastructure of American military power.

Compare this to any other defense contractor relationship in U.S. history. Lockheed Martin makes the F-35. Boeing makes the bombs. Palantir — which developed Maven Smart System — at least operates at arm’s length from the individuals setting national policy. Musk is not at arm’s length. He has operated inside this administration at the highest levels, and his companies have been woven through its security apparatus in ways that no previous contractor has approached.

Palantir CEO Alex Karp said last year that the company’s purpose was to “scare enemies, and on occasion, kill them.” It’s a blunt statement of mission. But Palantir doesn’t own the satellites carrying the targeting data. Musk does.

III. The Replacement

There is another company in this story. Its name appears in the court documents, and in the reporting around the Minab school strike, in ways that complicate any simple narrative about AI and warfare.

Anthropic.

Before Grok was integrated into Maven, Anthropic’s Claude was the AI model supporting the system. The Washington Post reported in March that the Pentagon used Claude and Maven Smart System while planning attacks on Iran — including to identify potential targets and provide precise locations.

Then Anthropic refused.

The company declined to reach an agreement with the Pentagon after the administration refused to guarantee that Claude would not be used for lethal autonomous weapons or the mass domestic surveillance of Americans. Those were Anthropic’s red lines. The administration wouldn’t cross them off the list. The deal fell through.

The Trump administration’s response was to designate Anthropic a ‘supply-chain risk to national security’ — a classification historically reserved for foreign adversaries. Then it turned to Musk.

The Minab school strike occurred on February 28, 2026 — the first day of Operation Epic Fury. A Tomahawk cruise missile hit the Shajareh Tayyebeh girls’ school in the Iranian town of Minab, killing at least 175 people. Most of them were children. It is the deadliest single civilian incident of the conflict.

Satellite imagery analyzed by Amnesty International showed the facility had been converted to a school long before the strike. The target data was outdated. Analysts have suggested a combination of AI-driven targeting and human error — failure to verify whether maps were current — contributed to the outcome.

When asked whether the reported use of AI in the Minab strike violated Anthropic’s red lines, CEO Dario Amodei said it did not necessarily do so — stressing that a human made the final decision. That answer is both technically defensible and ethically uncomfortable. The human in the loop approved a strike on a building that satellite imagery showed had been a school for years. The AI processed the intelligence. The human signed off.

Somewhere in that sequence, 175 people died.

IV. The Conflict That Isn’t Being Named

Let’s be precise about what is actually happening here, because the standard conflict-of-interest framing is insufficient.

A conflict of interest typically describes a situation where a person’s private financial interests might improperly influence their public duties. Musk selling rockets to the government while advising the government on policy is a conflict of interest. That’s been covered.

This is something different. This is vertical integration of the state itself.

Musk’s companies are not vendors competing for government contracts in an open market. They are the infrastructure. The satellites that carry the military’s communications. The AI that identifies the targets. The data centers that power the AI. The rockets that launch the satellites. And now, through the SpaceX acquisition of Cursor, the tools that the developers who build all of it use every day.

When the infrastructure of military power and the private interests of the world’s first trillionaire become structurally indistinguishable, the word ‘contractor’ no longer captures what’s happening.

Stanley’s testimony described xAI’s computing capacity as “as foundational to our modern defense posture as traditional munitions production.” That’s the Pentagon’s own framing. A private company’s data center, run by a man who also holds unprecedented access to the executive branch, is now officially equivalent to a weapons factory.

Senator Kirsten Gillibrand introduced legislation seeking to ban the use of large language models without human oversight in decisions involving the use of force or high-consequence actions. “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 wrote. The bill is pending. The strikes are not.

V. What This Changes

The SpaceX IPO priced the company at $2.5 trillion. The Cursor acquisition added $60 billion in developer market share. Grok’s integration into Maven Smart System made the AI infrastructure that underlies all of it a matter of “paramount national security.”

What that means in practice: the data centers can’t be shut down for environmental violations. The contracts can’t be terminated without compromising national security. The infrastructure can’t be audited the way a traditional contractor’s would be, because doing so would “severely impair” the Pentagon’s mission. Each legal, regulatory, or competitive pressure point has been neutralized by the same phrase.

This is not accidental. It is the architecture of a position that cannot be unwound without cost.

Meanwhile, Anthropic — the company that said no — has filed a confidential S-1 targeting a valuation of approximately $965 billion. Its revenue run-rate hit $47 billion in May, up from $9 billion at the end of last year. Its enterprise market share surpassed OpenAI’s for the first time in April. It is building the most commercially successful AI company in the world, in part, by refusing to be the infrastructure of a war.

Whether that refusal is principle or strategy — or both — is a question worth sitting with.

The company that said no is now worth nearly a trillion dollars and growing faster than any AI company in history. The company that said yes is now indistinguishable from the state it serves.

The Perplexity CEO said the SpaceX IPO would be a leading indicator for Anthropic and OpenAI’s listings. He was thinking about multiples and market sentiment. He probably wasn’t thinking about sworn testimony in an environmental lawsuit disclosing that one man’s AI helped pick 2,000 targets in four days.

But that’s the actual leading indicator. Not the stock price. Not the revenue multiple. Not the valuation.

The question the market hasn’t priced yet: what happens when the infrastructure of American military power belongs to one man, and that man’s data center can’t be regulated because shutting it down would compromise national security?

We don’t have an answer. We have a data center in Memphis, a school in Minab, and a sworn declaration that makes them part of the same sentence.

THE CHAOS TRANSLATOR • @currentlyted

Making complex systems legible.