Federal Appeals Court Upholds Pentagon Blacklist of Anthropic Over AI Safety Rules

Anthropic has argued that its strict internal AI safety constitution makes its models more secure for government deployment, not less. But on September 25, 2026, a federal appeals court ruled that the Pentagon was legally justified in viewing those exact safety constraints as a national security supply chain risk, upholding a blacklist that bars the Defense Department from using the company's technology.[1]

The appellate decision cements a standoff that began when the AI developer refused to waive its core safety guidelines to secure military contracts. The Trump administration's Defense Department subsequently labeled the company a supply chain risk, a designation the court has now affirmed as a lawful exercise of executive authority over military procurement.[1][3]

The core of the dispute centers on operational control. By enforcing an immutable "constitution" that restricts how its models can be deployed—particularly in combat, surveillance, or autonomous weapons scenarios—Anthropic maintains the ability to throttle or restrict outputs based on its own ethical framework.[2][4]

Government lawyers successfully argued that a vendor capable of unilaterally overriding military directives introduces unacceptable operational vulnerabilities into the defense supply chain. The appellate judges agreed, finding that the Pentagon's risk assessment properly accounted for the dangers of relying on software governed by private, non-waivable safety constraints.[5]

The ruling establishes a major legal precedent for how the federal government procures frontier artificial intelligence. It signals that defense agencies can legally disqualify AI vendors who refuse to cede full operational control of their models to the military, prioritizing chain-of-command reliability over corporate safety frameworks.[1][4]

Anthropic's designation as a supply chain risk effectively cuts the company off from lucrative Defense Department contracts, a sector that rival developers have aggressively pursued. The blacklist prevents the Pentagon from integrating Anthropic's Claude models into any defense systems or analytical pipelines.[5]

The decision also highlights a growing fracture between commercial AI safety researchers and national security imperatives. While developers increasingly rely on constitutional AI and strict usage policies to prevent model misuse, defense officials view those same guardrails as unacceptable limitations on battlefield utility.[2][3]

The court's affirmation leaves Anthropic with few avenues to challenge the designation short of an appeal to the Supreme Court. The immediate consequence is a bifurcated market where military AI infrastructure will likely be built exclusively by developers willing to bypass their own safety restrictions when the Pentagon requires it.[1][5]