A federal judge in San Francisco ruled that the Pentagon’s designation of Anthropic as a supply chain security risk violated the law. U.S. District Judge Rita Lin stated the government lacked sufficient rationale for the action.

Lin argued that the Department of Defense crossed a constitutional line by using the designation as payback for the company’s public criticism. The order noted that national security decisions warrant judicial deference, but the filings pointed to Anthropic’s media posture and pushback on military AI deployment as primary factors.

Legal Basis for Ruling

The judge wrote that neither the Constitution nor the invoked federal statute allows sweeping penalties based principally on Anthropic’s critique of the Administration’s views. The Pentagon claimed it could not trust Anthropic to ensure model integrity due to the company’s increasingly hostile manner through the press.

Background of Dispute

The dispute traces back to February, when negotiations over access to Anthropic’s Claude models fell apart. Anthropic sought commitments that its systems would not be used for full autonomous weapons or mass domestic surveillance. The department pressed for broad access for any lawful use.

Anthropic challenged the action in two courts: San Francisco and Washington, D.C. Lin’s ruling addressed the California matter. The D.C. litigation continues, leaving the company technically under the supply chain risk label until that case is resolved.

Company Response

An Anthropic spokesperson welcomed the court’s ruling that the designation was unlawful. The company stated it remains focused on working productively with the government to harness AI for national security so all Americans benefit from this technology.

Benzinga contacted the Pentagon for comment. The department acknowledged receipt of the request but did not respond before publication.

Anthropic launched a research preview of its Model Hardware Standard (MHS) on Thursday. The specification enables AI agents to discover, communicate with, and safely control physical devices such as laboratory equipment and robots.

The initiative targets a persistent integration challenge in manufacturing and scientific labs. Anthropic noted that connecting devices from different vendors often takes weeks or months due to proprietary programming interfaces. MHS introduces a standardized software driver acting as an intermediary, using basic commands like reading or writing temperature settings.

How It Works

The standard allows AI agents to operate multiple machines simultaneously, including microscopes, liquid handlers, and robotic arms. Users can describe safety limits in natural language, enabling the system to generate reference files detailing machine capabilities and constraints.

Once connected, devices are controlled via Anthropic’s Model Context Protocol. Agents can coordinate instruments, monitor results, and adjust parameters dynamically. Early testing showed Claude adjusting a laser and observing beam movement via camera until it learned the sequence, then converting that learning into a deterministic script for single-command alignment.

Industry Adoption

Several companies are testing or adding support for MHS:

• Amazon Web Services plans support through Strands Robots.
• Automata, Danaher, Doosan Robotics, MBF Bioscience, QIAGEN, Tecan, and Universal Robots are exploring integrations.

QIAGEN is testing MHS on its nucleic acid purification platform to help AI agents troubleshoot instrument problems. Tecan is adding support to its Fluent liquid-handling systems.

What the Numbers Show

The operational reality of current AI hardware control reveals a significant dependency on human oversight. While Anthropic demonstrated Claude’s ability to learn physical sequences, Genentech researchers had to intervene to clarify that foaming in protein samples was a physical issue rather than a software bug. This indicates that despite standardized communication protocols, AI agents currently lack sufficient contextual understanding of physical phenomena to operate fully autonomously without expert validation.

Limitations and Next Steps

Anthropic acknowledged that Claude’s understanding of the physical world remains imperfect. The technology currently works only with hardware possessing a programmable interface. The company plans to use the research preview for additional safety evaluations and is working with manufacturers to expand compatibility.