Anthropic Rebukes Tampering Claims: Says It Cannot Alter Deployed Claude AI For U.S. Military

In a court filing published Friday, a senior Anthropic executive confirmed that the company has no ability to modify its generative AI model Claude once the U.S. military launches and runs the system on its own infrastructure. The statement was issued to counter accusations from the Trump administration that the AI lab could tamper with its technology mid-military operation.

“Anthropic has never had the ability to cause Claude to stop working, alter its functionality, shut off access, or otherwise influence or imperil military operations,” wrote Thiyagu Ramasamy, Anthropic’s head of public sector. “Anthropic does not have the access required to disable the technology or alter the model’s behavior before or during ongoing operations.”

The Pentagon has been locked in a months-long dispute with the top AI research lab over rules for national security use of Anthropic’s technology, including what boundaries should govern that deployment. Earlier this month, Defense Secretary Pete Hegseth labeled Anthropic a supply chain security risk, a designation that will block the Department of Defense — and third-party contractors working with DoD — from using the company’s software over the coming months. Other federal agencies have also begun phasing out Claude for official work.

Anthropic has filed two constitutional challenges to the ban, and is seeking an emergency court order to overturn the designation. Even as the legal process moves forward, customers have already started canceling existing contracts with the company. A hearing for one of the lawsuits is scheduled for March 24 at the U.S. Federal District Court in San Francisco, and the judge could issue a ruling on a temporary reversal of the ban shortly after the proceeding.

In a filing earlier this week, government attorneys argued that the Department of Defense “is not required to tolerate the risk that critical military systems will be jeopardized at pivotal moments for national defense and active military operations.”

As WIRED has previously reported, the Pentagon has already used Claude for a range of tasks: analyzing large datasets, drafting internal memos, and assisting with the development of battle plans. The government’s core concern is that if Anthropic objects to how the military uses its AI, the company could disrupt active operations by cutting off access or pushing harmful unauthorized updates.

Ramasamy rejected that risk outright. “Anthropic does not maintain any back door or remote ‘kill switch,’” he wrote. “Anthropic personnel cannot, for example, log into a DoD system to modify or disable the models during an operation; the technology simply does not function that way.”

He added that any updates from Anthropic would only go live with explicit approval from both the U.S. government and its cloud service partner (widely known to be Amazon Web Services, which Ramasamy did not name directly). Ramasamy also confirmed that Anthropic cannot access the prompts or any other sensitive data that military users input into deployed Claude instances.

Anthropic leaders have repeatedly stressed in court filings that the company has no interest in holding veto power over U.S. military tactical decisions. Sarah Heck, Anthropic’s head of policy, wrote Friday that the company offered to formalize this guarantee in a contract proposal submitted March 4. According to the filing (which uses an alternative historical name for the Pentagon), the proposal stated: “For the avoidance of doubt, [Anthropic] understands that this license does not grant or confer any right to control or veto lawful Department of War operational decision‑making.”

Heck also noted that Anthropic was willing to accept contract language addressing the company’s own internal policy concerns about Claude being used to carry out lethal strikes without direct human oversight. Despite these concessions, negotiations between the two sides ultimately broke down.

For now, the Defense Department said in court filings that it “is taking additional measures to mitigate the supply chain risk” posed by Anthropic, by “working with third-party cloud service providers to ensure Anthropic leadership cannot make unilateral changes” to the Claude systems already in military use.