US Defense Secretary Pete Hegseth is pressuring Anthropic to provide the military with unrestricted access to its Claude artificial intelligence model, according to a report from Axios. Hegseth reportedly informed Anthropic CEO Dario Amodei that the company has until Friday to grant the Pentagon full access or face significant penalties.
The threats include invoking the Defense Production Act, severing existing contracts, and designating Anthropic as a supply chain risk. Such a designation would force other defense contractors to certify that they do not use Claude in their workflows. The Pentagon’s demand centers on the military’s reliance on Claude for its most sensitive operations. A defense official told Axios that the military requires Anthropic’s technology immediately because it is currently the only AI model used for highly classified work.
The official noted, “The only reason we’re still talking to these people is we need them and we need them now.” Claude has reportedly been utilized in specific military operations, including the raid targeting Nicolas Maduro in Venezuela. Anthropic has offered to apply its standard usage policies to Pentagon contracts but has drawn lines at certain applications. The company stated it would not allow its model to be used for the mass surveillance of American citizens or for autonomous weapons systems that operate without human oversight.
On the same day news broke regarding the Pentagon’s pressure campaign, Anthropic announced a significant overhaul of its safety protocols. The company revealed it is modifying its Responsible Scaling Policy (RSP), effectively lowering previously established safety guardrails. For years, Anthropic’s central promise to consumers and business partners was that it would halt the training of new AI models if specific safety benchmarks could not be guaranteed in advance.
This previous policy relied on “hard tripwires” designed to stop development immediately if certain risk thresholds were met. The updated policy replaces those strict red lines with a more flexible approach. Anthropic’s new framework relies on “Risk Reports” and “Frontier Safety Roadmaps” rather than automatic pauses in development. In a statement, Anthropic explained, “Two and a half years later, our honest assessment is that some parts of this theory of change have played out as we hoped, but others have not.”
The company argues that the shift is necessary to remain competitive in a rapidly evolving market. Jared Kaplan, Anthropic’s chief science officer, defended the decision in an interview with Time. Kaplan stated that unilaterally pausing development would be counterproductive while competitors continue to advance. “We felt that it wouldn’t actually help anyone for us to stop training AI models,” Kaplan said.
“We didn’t really feel, with the rapid advance of AI, that it made sense for us to make unilateral commitments… if competitors are blazing ahead.” Anthropic cited a “collective action problem” as a primary motivation for the policy change. The company argued that if one developer pauses for safety while others proceed without strong mitigations, the overall safety of the ecosystem could degrade.
According to the updated RSP, “The developers with the weakest protections would set the pace, and responsible developers would lose their ability to do safety research and advance the public benefit.” Anthropic also pointed to the United States’ broader anti-regulatory environment as a factor influencing its decision. Financial context underscores the competitive pressures facing the company.
In February, Anthropic secured $30 billion in new investments, bringing its total valuation to $380 billion. The company’s latest models, particularly in coding capabilities, have received widespread praise. However, this growth coincides with a shift in corporate ethos. While Anthropic’s previous safety commitments were a cornerstone of its brand, the new approach prioritizes continued training over hard stops.
Chris Painter, the director of the nonprofit organization METR, offered a mixed assessment of Anthropic’s new policy. Painter praised the increased emphasis on transparency but warned that the flexibility of the new RSP could lead to a “frog-boiling” effect. This phenomenon describes a scenario where safety standards are eroded incrementally through a series of rationalizations, eventually leading to a significant departure from original safety goals.
Neither Anthropic’s announcement regarding the RSP nor the reporting on the Pentagon’s demands explicitly connected the two events, though their timing on the same day has drawn attention to the intersection of government pressure and corporate safety policy.
