Defense Secretary sets deadline for Anthropic to grant military full AI access

2026-02-25

The BareStory

Defense Secretary Pete Hegseth met with Anthropic CEO Dario Amodei on Tuesday, setting a Friday deadline for the technology company to agree to the military's demands for access to its artificial intelligence models. The ultimatum follows a reported breakdown in negotiations regarding how the Department of Defense may utilize the company's software.

According to sources familiar with the meeting, Hegseth demanded a signed agreement granting the Pentagon full access to Anthropic's AI for all lawful activities. The company, however, has reportedly sought specific guardrails to prevent its technology from being used for mass surveillance or autonomous weapons systems. Amodei reportedly expressed concern that the AI is not yet reliable enough to avoid mistakes in lethal targeting decisions.

Defense officials countered that the military requires the same level of control over software as it possesses over hardware, such as aircraft. Officials also noted that mass surveillance is already illegal and stated the department is only seeking licenses for lawful operations. Sources indicated that if the company fails to comply, the government is considering invoking the Defense Production Act or designating Anthropic a "supply chain risk," a move that could preclude the firm from future government work.

Anthropic was awarded a $200 million contract last year and was the first AI company authorized to operate on classified military networks. However, a senior Pentagon official noted that competitor xAI has now also been cleared for classified use. An Anthropic spokesperson stated the company is continuing "good-faith conversations" regarding its support for the government's national security mission.

Left Perspective

Preventing Lethal Glitches
Checking Automated Warfare
Rejecting State Coercion

Right Perspective

Ensuring Operational Sovereignty
Trusting Legal Frameworks
Mobilizing Strategic Advantage

Left Perspective

• Preventing Lethal Glitches Prioritizing humanitarian safety, this camp views the hesitation of CEO Dario Amodei not as obstinance, but as a moral imperative to prevent algorithmic errors in "lethal targeting decisions." The logic holds that AI models are probabilistic rather than deterministic; deploying technology that is "not yet reliable enough" introduces unacceptable risks of civilian casualties. If a system cannot guarantee accuracy, ethical frameworks demand it remain fenced off from the kill chain.

• Checking Automated Warfare This perspective argues that "lawful activities" is an insufficient constraint without specific technical guardrails against "mass surveillance or autonomous weapons." The fear is that once full access is granted, the speed and scale of AI could outpace human oversight, turning software into an unaccountable force. By demanding hard limitations, proponents seek to ensure that decision-making power remains with human agents rather than black-box algorithms.

• Rejecting State Coercion Critics of the Defense Secretary’s ultimatum view the threat of invoking the "Defense Production Act" or applying a "supply chain risk" label as a dangerous overreach of state power against private innovation. This approach interprets "good-faith conversations" as a collaborative necessity, arguing that forcing compliance through existential threats to a company’s business undermines the partnership required to build safe, stable technology.

Right Perspective

• Ensuring Operational Sovereignty Prioritizing national security and the chain of command, this camp endorses the Pentagon's stance that the military must possess the same "control over software as it possesses over hardware." The logic dictates that a defense apparatus cannot effectively operate if its tools—whether aircraft or algorithms—come with corporate terms of service that override mission parameters. Full access is viewed not as a desire for abuse, but as a prerequisite for military readiness.

• Trusting Legal Frameworks This perspective dismisses corporate demands for extra guardrails, emphasizing that existing laws already make "mass surveillance" illegal and that the department is explicitly seeking licenses only for "lawful operations." Realists argue that the definition of permissible warfare belongs to the state and established law, not to private CEOs. Attempting to restrict the government’s capabilities implies that the private sector supersedes the legal authority of the Department of Defense.

• Mobilizing Strategic Advantage Viewing the $200 million contract and the clearance of competitor xAI as evidence of a competitive landscape, this side sees the ultimatum as a necessary move to secure "supply chain" reliability. If a vendor cannot commit to the "national security mission" due to internal hesitation, the state must pivot to partners who will. The threat of the Defense Production Act is interpreted here as a standard mechanism to align industrial output with urgent national defense needs.

How it may affect me

As a U.S. reader: If the military secures full access without the requested guardrails, U.S. defense operations may increasingly rely on AI systems that the manufacturer currently warns are not reliable enough for lethal targeting decisions. The potential invocation of the Defense Production Act could establish a precedent for how the federal government compels private technology firms to align their innovations with military requirements. While Defense officials assert that mass surveillance is illegal, the removal of technical restrictions requested by Anthropic could alter the capabilities available to the government for autonomous operations. Taxpayer-funded defense spending may shift to alternative competitors like xAI if the government designates current contractors as supply chain risks for failing to grant full software access.

Defense Secretary sets deadline for Anthropic to grant military full AI access

The BareStory

Left Perspective

Right Perspective

How it may affect me

Read the story at