The company has disputed the Pentagon’s statements that it can somehow still control Claude AI deployed in military networks
© Getty Images / NurPhoto / Samuel Boivin
AI developer Anthropic insists it has no backdoor or “kill switch” for its Claude AI once it is deployed in classified Pentagon military networks, according to a new court filing.
The
US military and the tech firm found themselves embroiled in a policy
dispute earlier this year, with the Pentagon insisting on using the
system for “all lawful military purposes,” while the company stood by its AI safeguards related to mass surveillance and fully autonomous weapons use.
The Pentagon ultimately ended its partnership with Anthropic, designating the tech firm a “supply chain risk,”
a rare label typically reserved for entities linked to Washington’s
foreign adversaries. The designation bars the company from not only
working with the US government directly but also any other contractors
from using its products as well.
In a new filing to a federal
appeals court in Washington, DC, Anthropic disputed the key US
administration claim that the firm still retained some degree of control
over Claude AI once it was deployed to classified systems and
effectively granted itself an “operational veto.” The firm said it has “no back door or remote kill switch,” while its “personnel cannot log into a department system to modify or disable a running model.”
The AI system supplied to the Pentagon comes as a “static” model, the company argued. Once it is deployed, it “does not degrade or change on its own, and Anthropic cannot push undisclosed or unsanctioned changes to a model after the department has deployed it.”
Anthropic was formally designated a “supply-chain risk to national security” on February 27, while US President Donald Trump accused it of being run by “leftwing nut jobs.” The company challenged the label in court, with the legal battle yielding thus far mixed results.
Earlier
this month, the DC court rejected Anthropic’s request for a pause on
the supply chain risk designation. In a parallel case in California,
however, a court sided with the company, temporarily blocking the
administration decision. With the split decision, Anthropic remains
barred from working with the Pentagon but can still continue its
partnerships with other agencies while the legal battle goes on.