A Battle Is Raging Between the War Machine and One of Its Profiteers

Anthropic AI is biting the hand that feeds them, and they may pay the price.

A battle is unfolding between the Pentagon and AI defense contractor Anthropic, which is best known for its Claude large language model (LLM). This week, the “Department of War” threatened to effectively blacklist the company from government contracts—or force them to comply—if they refuse to satisfy the military’s desire for a free-for-all on the technology.

This free-for-all reportedly entails the ability to mass surveil Americans and unrestricted uses of AI weaponry.

“Bright red lines”

In a post in January of this year, Anthropic CEO Dario Amodei declared some uses of AI “red lines.” He wrote that “using AI for domestic mass surveillance and mass propaganda—seem to me like bright red lines and entirely illegitimate.” He went on to claim that fully autonomous AI weapons and AI for use in strategic decision-making are a blurrier line to draw because national defense is a “legitimate” application for them.

However, he still expressed a desire for safety guardrails around such weapons. “My main fear is having too small a number of ‘fingers on the button,’ such that one or a handful of people could essentially operate a drone army without needing any other humans to cooperate to carry out their orders,” he wrote. In the post, he expressed concerns that AI tools could compromise democracy and allow people in power to behave as totalitarians. He also expressed concerns regarding what existing autocratic governments may do with these tools.

These concerns are reportedly driving the standoff between the “Department of War” and Anthropic. It’s worth noting that based on available reporting, much of this story relies on statements from anonymous Pentagon officials, making it difficult to truly ascertain what is happening behind closed doors. That said, Pentagon officials have reportedly claimed Anthropic’s concerns are unwarranted because the military only wants the technology for “lawful” activities. There is allegedly no need to worry about illegal surveillance, unjust or illegal use of weapons, or anything else. But even a cursory review of US history or current events shows the US government cannot be trusted to follow the Constitution and other laws.

In retaliation for Anthropic’s resistance, the Pentagon may deem the company a “supply chain risk,” which would effectively blacklist them from government contracting. On Thursday, the Pentagon reportedly took its first step in this process. Secretary of War Pete Hegseth also reportedly signaled he could use the 1950s-era Defense Production Act to force the company’s hand and give the military free rein over its tools.

Too little too late

While I, like Amodei, oppose using AI (or any technology) for mass surveillance and warfare, his opposition reads somewhat disingenuous; as this standoff has been unfolding, Anthropic announced a loosening of safety restrictions on their products. The company had previously pledged that it would not develop AI technologies beyond the point where it could guarantee such tools could be used safely. However, facing the possibility of falling behind other players in the AI race, they altered their own guardrails.

Further, despite Amodei’s concerns about mass surveillance, he apparently had no problem partnering with none other than Palantir (along with Amazon Web Services) to deliver Claude to the US military. Claude was the first LLM to receive clearance for classified use. Palantir has denied any role in helping the US government conduct domestic surveillance. However, previous reports spanning last year all the way back to whistleblower Edward Snowden’s 2013 releases suggest the company has no issue helping facilitate governments’ spy operations, whether they target Americans or others.

In light of all this, Amodei’s and Anthropic’s resistance to the worst uses of AI in government are arguably too little too late. He may justify Anthropic’s partnership with the military by citing national defense and protecting democracy, but the nature of government and power is to expand and encroach. Whatever misgivings Amodei may have about “red lines,” the conflict between his company and the Pentagon reflects this reality.

Case in point, the current flare-up of tensions between Anthropic and the Pentagon reportedly arose when the company discovered Anthropic tools were used in Trump’s regime change operation in Venezuela (Anthropic has denied harboring concerns about this). Regardless, the company clearly doesn’t get to pick and choose how the military applies their technology. That evidently (and unsurprisingly) went out the window when they signed up for military-industrial funding. And, predictably, the military is demanding unfettered access regardless of whether or not Amodei and others at his company consent.

An insidious history

Unfortunately, the history of the tech industry is steeped in collaboration with and dependence on the military. The Cold War saw the burgeoning technology industry profit and thrive off the United States’ government penchant for war, militarism, and eventually surveillance—a relationship that continues to this day. It is a staggering feat for engineers to find work in Silicon Valley that doesn’t eventually provide technology to the US government (and others) in one way, shape, or form. Indeed, Elon Musk’s xAI is happy to take Anthropic’s place. This week, xAI agreed to provide its Grok services for classified sectors of government. The Pentagon has moved forward with this arrangement despite Grok’s subpar security, reflecting Secretary Hesgeth’s growing track record of flouting safety precautions.

Predictably, those in positions of power believe everyone else must bend to their demands based on the construct of “authority.” Under Secretary of Defense for Research and Engineering Emil Michael even argued that Anthropic’s resistance to providing unmitigated AI access for the Pentagon was “not democratic.”

As of Friday, Amodei said the Pentagon’s threats wil not change his position, and Emil Michael called him a “liar” with a “God-complex.” This is comically hypocritical considering the Pentagon’s long track record of dishonesty and authoritarianism. While it remains unclear what either party will do, it is doubtful Anthropic will be able to successfully bite the violent hand that feeds them.