White House drafts guidance to restore Anthropic access as Pentagon legal fight continues

White House Prepares Path to Restore Anthropic Access

The White House is developing draft guidance that could enable federal agencies to sidestep Anthropic's supply chain risk designation and gain access to new AI models, including the advanced Mythos system 1

1

3

3

. According to Axios, a draft executive action currently under consideration could provide the Trump administration with a pathway to de-escalate its contentious dispute with Anthropic, with one source describing the White House efforts as a way to "save face and bring em back in" 3

3

.

This development marks a dramatic shift for an administration that previously treated the San Francisco-based AI company as a grave security threat. Earlier this month, White House Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent met with Anthropic CEO Dario Amodei for what both sides characterized as a productive introductory meeting 3

3

4

4

. The White House is now convening companies across various sectors this week to inform the potential executive action and establish best practices for deploying Mythos, including "table reads" of possible guidance that could walk back the Office of Management and Budget's directive prohibiting government use of Anthropic 3

3

.

Pentagon Dispute Over Autonomous Weapons and Domestic Surveillance

The conflict between Anthropic and the Pentagon originated when the startup refused to remove guardrails against using its Claude AI for autonomous weapons or domestic surveillance 1

1

. Anthropic insisted on banning its use for mass domestic surveillance or to develop fully autonomous weapons, declining to sign an agreement allowing the Pentagon to use Claude for "all lawful purposes" 3

3

. In response, the Pentagon designated the Claude-maker as a supply chain risk and canceled a $200 million contract 2

2

.

President Donald Trump previously declared on Truth Social that "The United States of America will never allow a radical left, woke company to dictate how our great military fights and wins wars!" 4

4

. However, his tone shifted last week when he said Anthropic was "shaping up" in the eyes of his administration, telling CNBC's "Squawk Box" that a deal with the Pentagon was "possible" because "We want the smartest people" 1

1

.

Mythos Drives Federal Demand Despite Legal Dispute

The reconciliation effort comes just weeks after Anthropic unveiled Mythos, its most advanced AI system to date, which experts say holds potentially unprecedented ability to identify cybersecurity vulnerabilities and devise ways to exploit them 1

1

. Agencies across the federal government are clamoring for access to Mythos even as the Pentagon battles Anthropic in court 3

3

. The National Security Agency is already using Mythos on classified networks, while government agencies including the Pentagon continue using Anthropic's models during the legal fight 3

3

.

According to the AI Security Institute, Mythos Preview became the first AI model to complete "The Last Ones," a 32-step corporate network attack simulation that typically requires humans 20 hours to finish 4

4

. Mozilla reported that using Mythos, the company identified and patched 271 software vulnerabilities in Firefox during testing, demonstrating the serious potential for the AI to both exploit current software and safeguard it from attackers 4

4

.

Anthropic Challenges Pentagon Claims in Court Filing

Anthropics told a U.S. Court of Appeals in Washington D.C. on Wednesday that it cannot manipulate Claude once deployed in classified Pentagon military networks, directly challenging the administration's supply chain risk designation 2

2

. In a 96-page filing, the company argues it has no visibility, technical ability, or any kind of "kill switch" for its technology once it's deployed, and that the Pentagon has the opportunity to test models before deployment 5

5

. The company contends the Pentagon is illegally retaliating against it by stigmatizing it with a designation meant to protect against sabotage of national security systems by foreign adversaries 2

2

.

Oral arguments are scheduled for May 19 before the appeals court 2

2

. Earlier this month, the D.C. court rejected Anthropic's request for an order blocking the Pentagon's actions, though the company prevailed in a separate parallel case in San Francisco federal court, which prompted the Trump administration to remove the stigmatizing labels from Anthropic according to court filings 2

2

. The split decision means Anthropic can't participate in new Pentagon contracts but can continue working with other federal agencies while the litigation plays out 5

5

.

Competing Interests and Uncertain Resolution

Multiple sources indicate that while key players at the Pentagon remain dug in on this issue, other stakeholders believe the fight has been counterproductive and are ready to find an offramp 1

1

3

3

. However, it remains unclear whether the steps under consideration would resolve the Pentagon fight or simply make it easier for other government agencies to work with Anthropic 3

3

.

The core dispute persists even if the supply chain risk designation is lifted. The Pentagon is currently operating on older terms of service that both sides view as overly restrictive and is not receiving the latest updates of the model 3

3

. Meanwhile, competitors OpenAI and Google have both signed agreements to let the Pentagon use their models under the "all lawful purposes" standard in classified settings, though both claim those deals respect the same two red lines Anthropic drew 3

3

. After the Pentagon canceled its contract with Anthropic, OpenAI struck a deal to provide its technology to the U.S. military 2

2

. Sources suggest it's possible the sides could end up right back in contentious negotiations 3

3

.