Anthropic is waging a court fight against the Pentagon to preserve bans on fully autonomous weapons and domestic mass surveillance. Defense Secretary Pete Hegseth demanded in January 2026 that all AI contracts be renegotiated to allow "any lawful use," The Verge reported on June 2. Anthropic objected — the only military AI contractor to set such limits. The dispute has drawn public attention to autonomous warfare. But it obscures a deeper reality: AI has been embedded in US military operations for 70 years.
"The systems we were talking about were not futuristic." Branka Marijan, a senior researcher at Project Ploughshares, attended a 2017 UN conference on lethal autonomous weapons in Geneva. She expected hypothetical discussions. That year was different.
The Pentagon had just launched Project Maven. Google had signed on. The mood shifted from speculative to acute, Marijan told The Verge.
The technology was no longer theoretical. It was here. Maven used AI to analyze drone surveillance footage.
Google employees revolted. About 4,000 signed a petition in April 2018 demanding the company exit the "business of war." Google chose not to renew the contract in mid-2018. Amazon and Microsoft quickly took over.
Palantir followed. Maven evolved into the Maven Smart System, or MSS. MSS now enables object detection, tracking, and large-scale surveillance analysis.
Claude, Anthropic's chatbot, was integrated into MSS in 2024 after the company loosened its military restrictions. Analysts could query the system about geographic areas or target types. Even that limited role increased efficiency.
The volume of targets grew, experts said. Human supervision became harder. "What we know about MSS is that it reduces the number of human beings in the targeting cycle — and that's actually by design," said Sarah Shoker, a senior research scholar at UC Berkeley and former lead of OpenAI's geopolitics team. Anthropic CEO Dario Amodei has drawn two red lines: no domestic mass surveillance and no weapons that identify, track, and kill without human involvement.
Those lines are "very narrow," said Maddy Batt, a legal fellow at Tech Justice Law. "Do not go far enough to protect human rights or to comply with international law." Amodei has expressed support for mass surveillance abroad. He has also said fully autonomous weapons "may prove critical for our national defense." He offered to help the "Department of War" speed up development. The Hegseth memo changed the calculus.
It explicitly prioritized speed over safety. "Must accept that the risks of not moving fast enough outweigh the risks of imperfect alignment," the memo stated. Hamza Chaudhry of the Future of Life Institute noted the memo "explicitly sets up a tradeoff and says that we favor speed." That approach carries lethal consequences. "Even without full autonomy, AI compresses kill chains to mere seconds so that humans are not actually making the assessments that international humanitarian law requires to prevent civilian harm," Batt said. The Pentagon designated Anthropic a military supply chain risk in March.
President Trump banned government agencies from using Claude. Relations have somewhat thawed since the release of Anthropic's cybersecurity-focused model Mythos. But the court battle continues.
OpenAI signed the terms Anthropic rejected. Seven other companies — Google, Microsoft, Amazon Web Services, Nvidia, Reflection, Oracle, and SpaceX — have since signed deals to deploy AI on classified Pentagon networks. Seventy years ago, a summer meeting between scientists in New Hampshire sparked the Defense Department's interest in AI.
The military has funded AI research ever since. The 2000s brought a surveillance revolution. Vast amounts of data could be parsed.
The 2010s delivered advanced facial recognition and machine vision. The momentum never stopped. "We've kind of crossed the rubicon while we pretend that we haven't," said Andrew Reddie, an associate research professor of public policy at UC Berkeley. DOD Directive 3000.09 sits at the center of the debate.
Iran Executes Two More Protesters as Families Search for Thousands Missing
Written in 2012, it defines fully autonomous weapons as systems that "once activated, can select and engage targets without further intervention by an operator." It requires humans to "exercise appropriate levels" of judgment. The definition is porous. Some missile defense systems like the Phalanx CIWS — a large automated gun defending naval vessels — have operated autonomously for decades.
They must react in milliseconds. A human in the loop would render them useless. The counterargument says these systems operate in fixed, defense-only environments.
They react. They do not decide. "The 'and' is doing a lot of work inside of that statute — we have systems that can decide and systems that can engage but you can't have a system that does both," Reddie said. "You cannot fight a war only in defense." Sorin Adam Matei, a Purdue University professor, was blunt. The Biden administration published a 2024 memorandum on AI and national security.
The policy remains in force under Trump. But the Pentagon's Chief Digital and Artificial Intelligence Office is undergoing a restructuring that isolates it further from the rest of the DOD. International efforts have stalled.
The Convention on Certain Conventional Weapons has met for over a decade. No binding agreement exists. There is not even an official international definition of "lethal autonomous weapon." "It's been over a decade, and there is really no agreement," Shoker said.
The public fight obscures quieter developments. Military AI integration hasn't slowed. Anthropic itself launched Claude Gov last summer — a product with loosened guardrails for classified materials.
The company is reportedly preparing an IPO this year at a $900 billion valuation. Investor pressure to turn a profit is intensifying. Employee dissent echoes the Google walkouts.
But Silicon Valley executives are pushing back, including using AI to identify leakers. Many tech workers fear for their jobs as AI replaces entry-level roles. Why It Matters: The negotiations between tech companies and the Pentagon will define the rules of engagement for decades.
If Anthropic loses its court fight, no major AI lab will have successfully imposed binding limits on military use. The technology that automates targeting is already deployed. The question is whether anyone remains accountable for what it does.
As Marijan put it: if humans are "just rubber-stamping decisions, what are the lines of accountability there — and what..."
Three developments now demand attention. Anthropic's court case will determine whether a private AI company can enforce usage limits on the Pentagon. The CDAO restructuring will shape how AI procurement and deployment decisions get made.
And international talks at the CCW continue without a legal instrument to govern autonomous weapons. Marijan said progress overall has "been very slow and we haven't seen concrete agreement, particularly among the major countries." The positions of China, Russia, and other military powers remain unaligned with any binding framework. The gap between what the technology can do and the laws that supposedly govern it widens every month. - The US military has embedded AI into warfare for 70 years, from early research funding to today's autonomous targeting systems. - Anthropic is the only major AI contractor to enforce limits on military use, but its red lines do not cover non-US persons or support for future autonomous weapons development. - Project Maven, launched in 2017, accelerated Silicon Valley's integration into warfare and evolved into MSS, a system now used in strikes on Iran and the capture of Venezuelan President Nicolás Maduro.
The fall of 2026 will bring inflection points. Anthropic's court case will either carve out a role for private limits on military AI or prove that government contracts overwrite all corporate red lines. The company's IPO prospectus will reveal how much military revenue underpins its valuation — and how much investors care about ethical restrictions.
The CCW will meet again in Geneva. The major military powers will restate their positions. No breakthrough is expected.
The twin forces driving this story — the Pentagon's push for speed and Silicon Valley's hunt for profits — align on one point. They both favor keeping humans away from the kill chain. The lines of accountability grow fainter with every contract signed.
Key Takeaways
— - The US military has embedded AI into warfare for 70 years, from early research funding to today's autonomous targeting systems.
— - Anthropic is the only major AI contractor to enforce limits on military use, but its red lines do not cover non-US persons or support for future autonomous weapons development.
— - Project Maven, launched in 2017, accelerated Silicon Valley's integration into warfare and evolved into MSS, a system now used in strikes on Iran and the capture of Venezuelan President Nicolás Maduro.
— - International regulation has stalled for over a decade with no binding agreement or even an official definition of lethal autonomous weapons.
Source: The Verge
