The Pentagon vs. Anthropic – The Dispatch

Speaking at a Medal of Honor ceremony on Monday, President Donald Trump said that the U.S. military operation in Iran was working to eliminate the country’s missiles, missile-making facilities, and naval targets. Earlier on Monday, Trump told the New York Post that he would consider placing U.S. troops on the ground in Iran “if they were necessary,” adding, “I don’t have the yips with respect to boots on the ground.” Similarly, Defense Secretary Pete Hegseth said during a press conference at the Pentagon on Monday that he would not rule out U.S. ground operations, but emphasized that the campaign would be “not endless.” Joint Chiefs of Staff Chairman Gen. Dan Caine noted that “this is not a single overnight operation,” adding that completion will require “difficult and gritty work.” To learn more about the Iran operation, read yesterday’s TMD.

Pakistani and Afghan forces exchanged fire for the fifth consecutive day on Monday, with the Taliban Defense Ministry claiming to have destroyed a Pakistani tank in southeastern Afghanistan after it fired shells. Taliban officials say their forces have taken 25 Pakistani strongholds, killing more than 100 soldiers. Pakistani fighter jets targeted Afghanistan’s Bagram Air Base on Sunday, and while Taliban officials claimed that Russian-made defense systems successfully repelled the Pakistani forces and that the facility didn’t sustain damage, satellite imagery obtained on Monday by the New York Times shows that an aircraft hangar and two large warehouses were destroyed. A Pakistani military official also told the outlet that Pakistan was targeting military equipment and supplies stored at the air base.

French President Emmanuel Macron announced on Monday that France was planning to expand its nuclear weapons arsenal, and that European allied nations would, for the first time, temporarily host its nuclear-armed aircraft for nuclear wargame missions, deterrence exercises, and “circumstantial deployments.” Macron said that France has begun preliminary talks with the U.K., Belgium, Denmark, Germany, Greece, the Netherlands, Poland, and Sweden on such drills, but emphasized that France would retain full authority over all nuclear-launch decisions. Macron said this is “forward deterrence” to counter what he described as “geopolitical upheaval fraught with ⁠risk.” He added, “To be free, one needs to be feared.”

A South Sudanese official on Monday said that insurgent militant fighters had attacked a village in the country’s northern region on Sunday, killing at least 169 people, including 90 civilians. Government soldiers and the insurgents battled for more than three hours. The attackers’ motive remains unclear to local officials. Authorities said that the death toll included women and children and that the figure may rise as more bodies are found. The United Nations shared in a statement that it was housing 1,000 people displaced from the attack at a nearby base, and identified the attackers as “unidentified armed youth.”

Ukrainian President Volodymyr Zelensky said on Monday that Ukraine will soon complete the technical requirements to advance negotiations on possible Ukrainian entrance to the European Union. Zelensky urged European leaders to schedule a specific date for Ukrainian accession, stating that such a move would pressure the Kremlin toward negotiating peace, though he later told reporters that “not everyone is ready to ⁠give Ukraine this opportunity.” Zelensky also said that Ukraine and Russia were still planning to hold in-person peace talks later this week, though he added Ukrainian officials were looking to change the location, possibly to Turkey or Switzerland. Before the U.S. and Israeli strikes on Iran this weekend, both sides had planned to hold talks this Thursday and Friday in Abu Dhabi. Kremlin spokesman Dmitry Peskov said that Russia was “open” to continuing talks.

Claude, Anthropic’s AI system, is likely playing a key role in America’s ongoing war with Iran. U.S. Central Command, which oversees the Middle East, uses Claude for planning, operations, battle simulations, and target identification, the Wall Street Journal confirmed Saturday. But the Trump administration has also just labeled Claude a threat to U.S. national security.

The U.S. attacks on Iran came just hours after President Donald Trump directed all U.S. government agencies—with a six-month transition period for the Department of Defense—to stop using Anthropic’s tools. Trump’s announcement capped a roughly weeklong saga in which a dispute between the Pentagon and Anthropic over how AI technology can be used in weapons systems and surveillance boiled over, ending Anthropic’s $200 million contract with the Department of Defense. Defense Secretary Pete Hegseth then designated the company a supply-chain risk, declaring that no military contractor, supplier, or partner may conduct any commercial activity with Anthropic.

The breakdown marks the first major collision between AI safety policy and government interests. But what led up to this? Do existing laws actually protect against the uses Anthropic tried to block? And what does the fallout mean for Anthropic and for the Pentagon’s relationship with Silicon Valley?

Anthropic, which has cultivated a reputation as the frontier AI company most committed to safety research and the ethical dimensions of its products, said negotiations with the government failed over a matter of principle: It refuses to allow its models to be used for deploying autonomous weapons or for mass surveillance of U.S. citizens. “We support all lawful uses of AI for national security aside from the two narrow exceptions above,” the company said in a press release on Friday, adding “we do not believe that today’s frontier AI models are reliable enough to be used in fully autonomous weapons” and “we believe that mass domestic surveillance of Americans constitutes a violation of fundamental rights.”

The White House and Pentagon maintain they have the sole authority to decide how to use AI tools. Jeremy Lewin, the acting undersecretary of state for foreign assistance, humanitarian affairs, and religious freedom, tweeted, “This isn’t about Anthropic or the specific conditions at issue. It’s about the broader premise that technology deeply embedded in our military must be under the exclusive control of our duly elected/appointed leaders.”

“They’re kind of talking past each other,” Bryan Clark, a senior fellow at the Hudson Institute, told TMD. “The Pentagon’s thinking, well, we don’t plan on using it for the things that you don’t want us to use it for, but we don’t want you to dictate how we employ the products that we buy from you. So it’s a little bit of a philosophical issue.”

Government officials have maintained that Anthropic’s objections were specious because U.S. law already addresses the company’s red lines. Pentagon spokesman Sean Parnell wrote on social media Friday that the Pentagon “has no interest in using AI to conduct mass surveillance of Americans (which is illegal) nor do we want to use AI to develop autonomous weapons that operate without human involvement.”

Strictly speaking, Parnell is correct, but the law is more flexible than the government’s framing suggests. The main legal authority governing the Pentagon’s use of AI weapons systems—Directive 3000.09—defines “lethal autonomous weapons” (LAWs) as a “weapon system that, once activated, can select and engage targets without further intervention by an operator.” LAWs aren’t banned, and there’s no public evidence the DOD has such weapons in its inventory, but the directive requires them to go through extensive review and have “human judgment over the use of force.”

At the moment, the military is mostly using AI for more mundane tasks. “The Pentagon is ultimately a giant bureaucracy that pushes a lot of paper, and so you can’t undersell how valuable something to speed that up is, to enable operations,” Lauren Kahn, a former Defense Department adviser who’s now a researcher at Georgetown University, told TMD. Jesse Kallman, the CEO of Danti—a geospatial search tool that can access data, internet search, and government data to aid operational planning—cited a concrete example: When major wildfires hit Chile earlier this year, Danti pulled together satellite imagery, social media, and weather data within seconds.

U.S. law also doesn’t quite rule out mass surveillance. “A lot of what you would colloquially think of as a search is not a search under the Fourth Amendment,” Alan Rozenshtein, a law professor at the University of Minnesota and a former Justice Department attorney, told TMD. User data, such as search history and algorithm preferences, collected and held by private companies, is not categorically protected by the Fourth Amendment (though that’s a subject of ongoing legal debate). Neither is CCTV footage collected in public places. Under the Foreign Intelligence Surveillance Act, the government can also acquire information from U.S. citizens collected incidentally during the surveillance of foreigners. AI-enabled search engines and image recognition tools could, in theory, massively expand the government’s ability to continuously monitor its own citizens.

The gap between Anthropic’s principles and legal reality seems to have doomed a last-ditch deal. The Atlantic reported Friday that Anthropic executives were optimistic about negotiating a new contract with the Pentagon, only to learn that the DOD intended to use Anthropic AI to analyze the bulk data of U.S. citizens, killing any chance of a deal. The company’s supply-chain risk designation swiftly followed. Interpersonal dynamics may have also soured the outcome. On Friday evening, Emil Michael, the Pentagon’s undersecretary for research and engineering, was reportedly on a call with Anthropic executives just before the deadline and asked to speak with Anthropic CEO Dario Amodei directly. Michael was told the CEO was in another meeting and unavailable.

As Anthropic’s deal with the government collapsed, its chief competitor, OpenAI, was crafting its own. On Friday, OpenAI announced it had agreed to a deal with the DOD to use its AI models in classified settings, and, on its website, OpenAI also took the unusual step of posting the “relevant” parts of its contract with the government, maintaining it upheld the same safety and usage red lines as Anthropic.

The published excerpts committed to human control over autonomous weapons and Fourth Amendment compliance for private data. OpenAI maintains its contract “provides better guarantees and more responsible safeguards than earlier agreements, including Anthropic’s,” in part because AI processing will happen in centralized data centers rather than on battlefield hardware, OpenAI’s own safety systems will remain active, and cleared OpenAI employees will have ongoing oversight of how models are used.

“OpenAI’s story is that Anthropic was wrong to try to rely on contractual requirements,” Charlie Bullock, a senior research fellow at the Institute for Law and AI, told TMD. “Instead of that, they’re going to rely on technical safeguards and having OpenAI employees in the loop.” But he noted that if the DOD then identified a lawful use of AI that violated OpenAI’s own safeguards, “it’s not clear whether OpenAI would be contractually obligated to comply.” It’s also unclear, Bullock and Rozenshtein both noted, whether a change in U.S. law would also change the terms of the contract.

On Monday night, OpenAI CEO Sam Altman said the deal would be amended to explicitly ban domestic surveillance of U.S. citizens—including through commercially acquired personal data—and that DOD intelligence agencies would be excluded from using OpenAI models unless the contract was renegotiated. Bullock noted that OpenAI’s deal appeared to place the same restrictions on AI use that Anthropic had demanded, raising the question of why Anthropic’s stipulations were deemed a supply-chain risk at all.

The government may also be exaggerating the effect of its penalty on Anthropic. “Effective immediately, no contractor, supplier, or partner that does business with the United States military may conduct any commercial activity with Anthropic,” Hegseth claimed in his tweet announcing the move. If taken at face value, this would bar Anthropic from conducting business with companies such as Amazon, which is simultaneously a government contractor, an Anthropic investor, and its primary cloud infrastructure provider. Anthropic disputed Hegseth’s assertion, claiming that Pentagon restrictions “can only extend to the use of Claude as part of Department of War contracts.”

“Hegseth’s statement is sort of wildly overstating what the law allows them to do,” Bullock said, noting that there also wasn’t any evidence the government had conducted the legally required investigation to designate the company as a supply-chain risk. “I think Anthropic has a strong legal case here,” he said.

The risk to Anthropic’s business is mixed. When companies such as the Chinese telecoms giant Huawei have been slapped with the supply-chain risk designation in the past, out of an abundance of caution, other companies “cut their technology out of their entire stack,” Bullock said. But Anthropic’s stance could be viewed similarly to Apple’s refusal in 2016 to build an encryption back door for the FBI, which lent weight to the brand’s claim to be the secure, private phone company. Anthropic may have lost a $200 million contract, but the company is on pace for $14 billion in annual revenue, and this news cycle has driven a dramatic increase in general consumer interest in Claude. In app downloads, this incident has seemingly been more effective advertising than its no-ads-in-AI Super Bowl commercial.

But no matter the legal outcome, the Pentagon’s apparent attempt to cripple Anthropic’s business jeopardizes the fragile relationship between Silicon Valley and the U.S. government.

And those who do business with the DOD are certainly paying attention. Designating a U.S. company as a supply-chain risk would be “highly unprecedented,” said Kallman. “It’s highly concerning.” Rozenshtein was even more pointed: “This should be understood as the first chapter in the nationalization of the American AI industry,” he said.

Fox News: FBI Files Reveal New D.B. Cooper Suspect—Maine Pilot Once Investigated in Skyjacking Mystery

New York Times: He Is Charged With Murder. He Could Become a Sheriff.

The Guardian: Private Jet Prices Soar As Wealthy Scramble To Leave Dubai