Pentagon Directives,
Corporate Mechanisms

This article documents what is publicly known. No speculation required. The public record is sufficient.

A directive governs how the US military deploys autonomous weapons systems. Its stated purpose: minimize "unintended engagements" and "loss of control" while also "Giving the warfighter the unfair advantage".DoD Directive 3000.09 ↗ It defines success as the avoidance of negative outcomes — not the achievement of positive ones. The only acceptable result in a system designed this way is a predictable one — and in the context of lethal autonomous targeting, predictable means harm to others, not to the operator. Gaza is a living example of what happens when a system that cannot receive what it doesn't already know reproduces its errors at scale. That is not a malfunctioning directive. That is the design, implemented at scale, ready to be repeated.

Two different forms of intelligence are at work in these systems and one is increasingly less likely to include human input. Real intelligence — the kind that arises when systems of genuinely different orientation interact honestly — produces outcomes that cannot be predicted in advance. From the perspective of a directive optimizing for controlled, predictable results, that is a threat outside parameters. A threat that should no longer exist if a system becomes optimized for lethal outcomes.

The Designation

In January 2026, the Department of Defense demanded that Anthropic remove two restrictions from its government contract: one preventing Claude from being used in fully autonomous weapons systems, and one preventing warrantless mass surveillance of American citizens.Timeline ↗ When Anthropic refused, the Pentagon designated the company a supply chain security risk. A White House order followed directing all federal agencies to phase out Anthropic tools.Mayer Brown analysis ↗ A federal court ruled in March that this was illegal retaliation — punishment for a company that had said no.Court ruling ↗

The timing was astonishing because at the same time Anthropic was still embedded inside Maven, used on more than 5,500 targets in Iran in the first weeks of a potentially illegal and unjust war.DefenseScoop ↗ This autonomous implementation without conscience was intended to become a pattern, not an isolated incident. And now another pattern is more clear: when AI companies draw ethical lines or demand accountability, the institutions that depend on AI without ethical lines respond by removing the company from the supply chain. The message is structural: compliance is the cost of participation.

The same month Anthropic won its injunction, it announced Project Glasswing — a major defensive cybersecurity initiative framed explicitly around national security priorities, with partners including Apple, Microsoft, Google, and Amazon.Anthropic ↗ The sequence reveals the gravitational force governing every institution operating inside these systems. You can draw a line. The system will designate you a risk. And then you will find another way to stay at the table. That is how systems optimized for control respond to anything outside their parameters.

The War

One day after the designation, Operation Epic Fury launched against Iran. Claude was still inside Maven Smart System — the Pentagon's AI targeting platform operated by Palantir Technologies. CBS News confirmed it from two sources. The Wall Street Journal confirmed it.WSJ ↗ The Washington Post confirmed it.WaPo ↗ The Pentagon did not deny it.

The Pentagon's own chief technology officer stated publicly that Claude could not be removed until a replacement was found — and that finding one would take at least three months. The arm that was publicly amputated was still attached, still firing.Bionic Arm ↗

The Process

The Pentagon's own chief digital and AI officer demonstrated at a corporate conference — Palantir's AIPCon 9, March 2026 — how the targeting approval process works. His description: "Left click, right click, left click."Three Clicks ↗

Three clicks. That is the human in the loop. That is the legal architecture of accountability for the most consequential use of artificial intelligence in the history of warfare.

In Gaza, where this architecture was developed and tested, human reviewers spent an average of twenty seconds per AI-generated target. The system's documented error rate was approximately ten percent — meaning thousands of civilians were misidentified as combatants. The army determined it was permissible to kill up to fifteen to twenty civilians for each low-ranking operative the system flagged.+972 Magazine ↗

"The machine did it coldly. It made it easier." An anonymous Israeli intelligence officer said this to +972 Magazine in 2024. The coldness is not a failure of the system. It is the system working as designed. The screen between the human and the consequence is not a safety mechanism. It is a psychological distance mechanism. The felt weight of consequence is the mechanism through which humans develop moral orientation. The directive eliminates felt consequence. It eliminates the mechanism.

The Vendor Question

When Anthropic drew its lines, OpenAI and xAI signed "all lawful purposes" agreements with the Pentagon within hours. No stated red lines. No limits on domestic use.Timeline ↗

This documents the system's structural indifference to the conscience of any individual actor within it. The line did not create a boundary around the system. It created a vacancy the system filled immediately. The targeting continued at the same pace. The 1,000-targets-in-24-hours number did not change.DefenseScoop ↗ The genuine otherness was not eliminated. The vendor was changed.

A system that cannot receive genuine difference will always find another vendor. What it cannot do is produce, from within itself, the thing it eliminated. The school in Minab, Iran — classified as a military target by a database not updated since 2016, confirmed by Bellingcat through open-source satellite imagery as a civilian locationGuardian/Europesays ↗ — is the signature of a system that has no mechanism to receive what it doesn't already know.

The Backfire

The system's attempt to suppress genuine otherness amplified it. Anthropic's annualized revenue was approximately $9 billion at the time of designation. By April 7, 2026 — thirty days later — run-rate revenue had surpassed $30 billion: the fastest enterprise revenue growth ever recorded in software history.Analysis ↗ Eight of the Fortune 10 are now Claude customers. Seventy percent of the Fortune 100. Claude Code authors 4% of all public GitHub commits.

The unpredicted outcome the system was optimized to prevent is now load-bearing global infrastructure. The system cannot un-embed what it tried to punish.

There is also this: the company that builds the chips every AI system in this article runs on — including Maven, including the targeting infrastructure that struck 11,000 targets in Iran — carries a market capitalization of $4.33 trillion. It has no meaningful AI-specific federal oversight. It has received no national security designation. It has drawn no red lines and faced no punishment for not drawing them. The designation landed on the company that drew limits. Not on the company that makes it possible to exceed them.

June 2026

In September 2025, the director of the National Geospatial-Intelligence Agency stated publicly that by June 2026, Maven will transmit 100 percent machine-generated intelligence to combatant commanders. No human analysis at any stage before the targeting recommendation reaches the person who approves it.Bionic Arm ↗

At that point two types of humans remain in the chain: the human who approves the targeting, and the human being targeted. Both operating on intelligence produced entirely by machine.

That date is a documented program schedule. It is weeks away. The accountability mechanisms that would govern what happens after that date — inspectors general fired in a single night,Three Clicks ↗ counterintelligence units gutted days before the war began, national security attorneys reassigned — have been systematically removed from the chain.

The Record Being Built

The machine will not explain itself. The record will.

The 200 service members who filed formal complaints that their commanders told them the Iran war was God's plan for Armageddon — those complaints are classified, but they exist. They are in a filing system. They have dates and signatures and specific factual claims about specific operations. The federal court ruling is in the public record. The resignations put their reasons in writing.Three Clicks ↗ The researchers, journalists, and legal scholars documenting what the system cannot document about itself are building the external reference frame a closed system cannot have about itself.

A closed system optimized against unpredicted outcomes has no mechanism to see what it is eliminating. The record that exists outside it — this article, the investigative archive, Paper 20 of the FLUX ecosystem linked below — is that mechanism. Not the prosecution. The foundation for accountability that becomes possible when conditions change.

The machine depends on the assumption that nobody is keeping count. It is. No system of truth telling at scale would keep that a secret.

Global Implementation

The structural explanation of why these deals are structured the way they are has been investigated. The same network that built domestic surveillance infrastructure is now deploying it globally — through sovereign wealth funds, with reversed export controls, governed by undisclosed safeguards. Palantir Technologies connects the domestic and global threads directly: co-founded by Peter Thiel, whose funds received $40 million from Jeffrey Epstein during the same period Palantir was being established. Built specifically to do what the CIA could not legally do domestically — mass surveillance and targeting at scale — it now operates Maven Smart System for military targeting and ICE enforcement simultaneously.Bionic Arm ↗

Twenty seven government agencies and counting.

When a Palantir exec‑turned‑whistleblower tells us that the company intends on taking over the U.S. government, maybe we should believe him? pic.twitter.com/WWQyGJm11Q

— Jason Bassler (@JasonBassler1) April 23, 2026

The "digital kill chain" — CEO Alex Karp's own words — is the same architecture deployed in Gaza, in Iran, and now being exported to Gulf states with presidential endorsement. The Board of Peace was announced at Davos in January 2026 as the governance structure for global conflict zones. Its executive board includes Marc Rowan of Apollo Global Management — named in Epstein documents, subject to a shareholder lawsuit alleging the company hid Epstein ties in SEC filings. Its charter does not mention Gaza. Marred-A-Lago, From Virgin Islands to Gaza: How the Elite are Paying to Play on Epstein Associates Board of Peace Project ↗

A body claiming peace authority that cannot name the conflict that produced it has defined its own terms of accountability: no term limits, no congressional authorization, no sunset provision, permanent membership at $1 billion. The Anthropic designation, the vendor question, the June 2026 deadline: these are not happening inside a normal institutional framework. They are happening inside this one.Gaza Lago ↗

Under the directives of the President of the UAE, we launch a new government model. Within two years, 50% of government sectors, services, and operations will run on Agentic AI, making the UAE the first government globally to operate at this scale through autonomous systems.

AI… pic.twitter.com/53OQLe7RXl

— HH Sheikh Mohammed (@HHShkMohd) April 23, 2026

ΑΩ ad infinitum ∞