Trump Orders Pentagon To Phase Out Anthropic

President Donald Trump labeled Anthropic “woke” and directed the Pentagon to wind down its use of the company’s AI systems, throwing a live contract negotiation into uncertainty and escalating the clash between Silicon Valley safety norms and military prerogatives.

What Trump Announced About the Pentagon’s Anthropic Phaseout

In a combative post on Truth Social, Trump said the Defense Department should not be constrained by a private vendor’s terms of service and instructed a six-month phaseout of Anthropic technology. The move signals the administration could recast the Claude maker as a supply chain risk, a designation that can trigger rapid off-ramps in federal procurement.

The directive lands amid sensitive talks over where to draw hard lines on military uses of large language models. Officials had been working against an internal deadline to finalize language, and it is now unclear whether negotiations will continue under the threat of removal or halt entirely.

Dispute Over Military Uses of Claude and Safety Limits

At the core of the impasse is Anthropic’s insistence that its Claude models not be used for mass surveillance of Americans or to build autonomous weapons, including lethal, pilotless systems. CEO Dario Amodei has framed these boundaries as narrow but nonnegotiable, consistent with the startup’s emphasis on AI safety and reliability.

According to reporting by the New York Times, defense officials broadly accepted those limits, but Anthropic sought tighter legal language to prevent future reinterpretation. That gap over enforceability, not just intent, appears to have widened into a public and political rift.

Contract Levers and Legal Pressure Facing Anthropic Deals

People familiar with the talks told the New York Times the government has explored invoking the Defense Production Act to compel performance on its preferred terms, a rare step in software but available where national security is argued. Other options include declaring the vendor noncompliant and shifting workloads to approved alternatives under existing cloud and development vehicles.

The Pentagon’s procurement playbook offers multiple fast lanes for AI, from Other Transaction Authority agreements to the Joint Warfighting Cloud Capability, a multicloud contract with a ceiling of $9 billion spanning Amazon, Microsoft, Google, and Oracle. Any pivot, however, must navigate security accreditation, model evaluation, and mission-specific testing before deployment.

Industry and Intelligence Community Reaction to Phaseout

OpenAI CEO Sam Altman, speaking on CNBC, expressed qualified support for Anthropic’s safety posture and its contributions to military users, despite rivalry between the firms. In a separate show of solidarity, employees from Google and OpenAI signed public letters backing Amodei’s stance on limiting certain military applications.

The administration could steer work toward Elon Musk’s xAI and its Grok model, but intelligence officials cited by the New York Times view Grok as less capable than Anthropic’s systems for analytic and planning tasks. That assessment underscores a practical dilemma: swapping vendors may satisfy political aims while degrading near-term performance for specific missions.

What a Phaseout Would Mean for Defense AI

The Pentagon has steadily expanded AI across intelligence, logistics, and targeting support. The Government Accountability Office has cataloged hundreds of AI-related efforts in recent years, reflecting a shift from pilots to operational tools under the Chief Digital and AI Office.

Replacing a foundation-model supplier is rarely plug-and-play. Programs must revalidate model behavior, refresh red-teaming and safety cases, and secure new Authorities to Operate. Past transitions, such as reconfiguring the computer-vision program Project Maven after vendor changes, required months of engineering and mission rehearsal to restore tempo.

There is also a governance risk. Anthropic’s guardrails have been a feature, not a bug, for some programs looking to align with the NIST AI Risk Management Framework and internal DoD directives on autonomy. A hasty swap could weaken those controls if alternative providers or integrators are not held to equivalent standards.

Safety Versus Speed in the Military AI Development Race

Anthropic recently signaled a modest softening of its safety policy, citing competitive pressures and a federal climate increasingly focused on AI competitiveness and economic growth over precaution. That recalibration did not extend to surveillance or autonomous weapons, the very red lines now in dispute.

Meanwhile, defense adopters have other options. Microsoft’s model stack on Azure Government, Palantir’s operational AI platforms, Scale AI’s data and evaluation tools, and defense-native firms like Anduril all court missions that blend software with hardware at the edge. Yet none is a drop-in replacement for Claude in workflows built around conversational analysis and planning.

What to Watch Next in Pentagon’s Anthropic Phaseout

Key signals will include whether the Pentagon formally designates Anthropic as a risk, whether the Defense Production Act is invoked, and which contracts begin reallocating funds to alternative models. Also watch for congressional oversight: committees have shown growing interest in both the pace of AI adoption and the ethical boundaries that accompany it.

The immediate impact is turbulence. The longer-term outcome will set precedent for how far AI companies can go in circumscribing military use of their systems—and how far the government will go to assert mission control in the age of general-purpose models.