AI company Anthropic said it cannot accept the Pentagon’s “best and final” offer to resolve a dispute over limits it has placed on how the U.S. military uses its AI models. The conflict has gotten uglier with just hours left before a Friday deadline by which Anthropic could be banned from doing business with companies that do business with the U.S. military if it doesn’t comply with the Pentagon’s demands.
Pentagon officials have publicly questioned the character of Anthropic CEO Dario Amodei. Meanwhile, employees at competing AI institutes have signed an open letter supporting Anthropic’s position. Axios reports that OpenAI CEO Sam Altman told employees in a memo on Thursday that OpenAI intends to impose similar restrictions on autonomous weapons and mass surveillance as Anthropic, which is in talks to extend the use of ChatGPT, currently available to the military for unclassified use cases, into more sensitive areas.
The battle between humanity and the Pentagon now threatens to escalate into an industry-wide revolt among technology workers at AI companies over how the AI systems they are building are used in the military. On Thursday, more than 100 Google employees sent a letter to Jeff Dean, the company’s chief scientist, asking for similar restrictions on how the U.S. military uses Google’s Gemini AI model. New York Times.
Amodei released a lengthy statement Thursday explaining why the company believes there should be limits on the use of its AI technology for autonomous weapons and mass surveillance. These are two areas where Anthropic currently limits the military’s use of the model, both in terms of contract and in the safeguards it has built directly into the Claude model. The Pentagon wants those restrictions lifted and Anthropic agrees to allow the U.S. military to use its models “for any lawful purpose.”
Amodei said in a statement that Frontier AI systems are “not reliable enough to power fully autonomous weapons,” and without proper oversight, “we cannot trust the critical judgment that our highly trained, specialized forces make every day.” When it comes to surveillance, he argued, powerful AI can now stitch together otherwise innocuous public data, such as location records, browsing history, and social connections, to create a comprehensive picture of the lives of Americans at scale.
U.S. Undersecretary of the Army Emile Michael called Amodei a “liar” with a “God complex” and accused the CEO of wanting “personal control over the U.S. military” in a post on social media platform X. In another post, Michael characterized Anthropic’s Claude Constitution, an internal document that outlines the values and principles the company builds into its AI, as a corporate conspiracy to “impose corporate law on Americans.”
The Pentagon has given Anthropic until 5:01 p.m. Friday to lift contractual restrictions it objects to, or the $200 million contract with the U.S. military could be canceled or, in more extreme cases, labeled a “supply chain risk,” effectively barring companies that do business with the military from using Anthropic’s technology.
This type of action is typically limited to foreign adversaries, such as China’s Huawei and Russian cybersecurity firm Kaspersky.
“Using this system against domestic companies because they don’t want to compromise these kinds of principles is really escalating and unprecedented,” said Sean O Hegiatai, a research professor at the Center for the Future of Intelligence. luck.
The Army Department has also threatened to invoke the Cold War-era Defense Production Act, which the government says is essential to national security and is using it to force Anthropic to hand over an unrestricted version of the Claude. If the Pentagon continues down this path, it would use emergency-only powers to resolve contract disputes in peacetime. There is some precedent for this. The Biden administration also invoked the DPA in 2023 to force frontier AI labs to hand over information about the safety of their AI models. But forcing companies to produce products rather than simply providing information comes closer to nationalizing big technology companies.
“When people are effectively forced to authorize technology to be used in a way that even they themselves say they can’t trust in a life-or-death situation like a battlefield, that sets a very dangerous precedent,” Ó Heigiatai said.
The War Department has publicly stated that it has no intention of conducting mass surveillance or removing humans from determining weapons targets, but the dispute may come down to how each side actually defines “autonomy” or “surveillance.” Representatives from the ministry did not respond to requests for comment. luck.
An Anthropic spokesperson said: luck The company said it continues to “work in good faith” with the Department of the Army. However, a spokesperson said the contract text received overnight showed “virtually no progress” on core issues. The new language, which was “framed as a compromise,” was “combined with legal language that allowed these safeguards to be freely ignored,” they said. Amodei said the Army Department’s threats are “inherently contradictory,” saying, “One side labels us a national security risk and the other label Claude as essential to national security.”
Anthropic has earned praise from some for its resolute stance. Lawrence Lessig, a Harvard law professor, praised the company’s statement as a “beautiful act of integrity and principle” and called it “incredibly rare in this day and age.”
Rivals OpenAI and xAI have agreed to a Pentagon contract that allows them to use their models for any lawful purpose, and xAI is going further by also agreeing to deploy its systems in some classified settings. But more than 330 current employees of competing labs, Google DeepMind and OpenAI, also released an open letter in support of Anthropic, urging their own leadership to follow the company’s lead. “They are trying to divide each company out of fear that the other will give in,” the letter said. “That strategy will only work if none of us knows the position of other countries.” Signatories included senior researchers from both companies, as well as anonymous and anonymous researchers.
Ó Heigiatai said the consequences of this debate could extend far beyond anthropology itself. “If the Department of Defense moves forward with this issue, it will establish a precedent that is not good for the independence and ability of these companies to uphold ethical standards,” he said.
