Taft’s prior alert addressed the Trump administration’s directives barring federal contractors from using Anthropic and its Claude platform in performing federal government contracts and subcontracts. On March 26, the U.S. District Court for the Northern District of California entered a preliminary injunction against the government’s enforcement of those directives. Judge Rita F. Lin concluded that Anthropic had demonstrated a likelihood of success on the merits of its claims, and that it was suffering irreparable harm from the challenged actions pending final resolution of the litigation.

Background

The dispute arose from Anthropic’s refusal to permit unrestricted government use of Claude for mass surveillance of Americans and for lethal autonomous weapons systems, applications that Anthropic maintained the technology was not yet safely capable of supporting. The government’s response extended beyond simply selecting alternative vendors. The President directed all federal agencies to immediately cease use of Anthropic’s products. Secretary Hegseth directed the Department of Defense to designate Anthropic a supply chain risk to national security under 10 U.S.C. § 3252, and further ordered that no military contractor, supplier, or partner could conduct commercial activity with Anthropic.

The Court’s Analysis

The court found that the record supported a plausible inference that the challenged actions were undertaken not to address legitimate operational concerns, but to penalize Anthropic for publicly disclosing its position in the contracting dispute. The Department of Defense’s own internal memorandum identified Anthropic’s engagement with the press as the basis upon which its risk level escalated. The court therefore concluded that the government’s actions far exceeded what any operational concern would require and was consistent with viewpoint-based retaliation prohibited by the First Amendment.

The court also found that both the Hegseth Directive and the Supply Chain Designation were likely in excess of statutory authority, contrary to law, and arbitrary and capricious. Section 3252 was designed to address covert acts of sabotage and subversion of national security systems, not overt positions taken by vendors during commercial negotiations. The congressional notification letters required by the statute contained no discussion of whether less intrusive measures were considered, as the statute expressly requires. The court further noted that the administrative record supporting the designation was generated almost entirely within a two-day window, and that the day after the designation was finalized, the Department of Defense’s Under Secretary informed Anthropic’s Chief Executive that the parties remained “very close” on contract terms.

Implications for Contractors

The preliminary injunction suspends enforcement of the challenged actions while the litigation proceeds. Contractors should nonetheless approach this development with measured caution. The government has been granted a seven-day administrative stay to seek emergency relief from the Ninth Circuit, and further appellate proceedings are anticipated. The injunction does not require the government to procure or use Anthropic’s products; it prohibits only active enforcement of the ban and the associated contractor blacklist.

For contractors managing their AI procurement strategies, the court’s analysis under Section 3252 is of particular significance. If the court’s reasoning is sustained on appeal, the supply chain risk designation process under that statute will be understood to have a narrower scope than the government has recently asserted, which would limit the executive’s ability to employ that mechanism as leverage against vendors engaged in public policy disputes with the Administration.

Taft will continue to monitor this matter and will provide further guidance as appellate proceedings develop. Questions regarding the impact of this decision on contracting postures or AI strategy should be directed to Taft’s Government Contracts or Privacy, Security, & AI teams.