Anthropic Refuses Pentagon Request to Strip AI Safety Guardrails

Published on 28 February, 2026

Anthropic has publicly refused a directive from the US Department of War to remove safety guardrails from its artificial intelligence systems. The company argues that current AI technology lacks the reliability required for unrestricted military applications and that complying with the demand would endanger American civilians and military personnel.


The dispute centers on a Pentagon ultimatum demanding unrestricted military use of Anthropic's Claude model. The Department of War threatened to cancel existing contracts and impose penalties if the AI firm did not comply. In a statement released Thursday, CEO Dario Amodei confirmed the company would not change its stance, emphasizing that private firms should not make military decisions but must draw the line at uses that undermine democratic values.


Specific Objections to Military Use


Amodei identified two critical areas where the technology is currently unsafe: mass domestic surveillance and fully autonomous weapons. Regarding surveillance, the CEO noted that modern AI can automatically construct a comprehensive portrait of any individual's life at a massive scale. He suggested this capability is currently outpacing legal frameworks.


Regarding autonomous weaponry, Amodei argued that frontier AI systems are simply too unreliable to power such tools without human oversight. He expressed concern that removing guardrails would put American warfighters and civilians at risk, noting that AI cannot yet exercise the critical judgment of trained professional troops.


A Standoff with the Pentagon


The statement highlights a contradiction in the Pentagon's position. The military has threatened to label Anthropic a national security threat for refusing to remove guardrails, while simultaneously arguing that removing those very restrictions is necessary for national security. Despite the threats of sanctions and contract cancellations, Anthropic has offered to collaborate on R&D to improve system reliability rather than disabling safety features.


The situation is heading toward a confrontation with Secretary of War Pete Hegseth, who has set a Friday deadline for Anthropic to acquiesce to the Pentagon's terms.

Comments

Leave a comment