Detailed Analysis
Federal agency staffers are continuing to use Anthropic's Claude AI tools weeks after President Trump issued a directive ordering the immediate cessation of Claude's use across U.S. government agencies, according to reporting by Information Security Media Group. Employees from multiple federal agencies confirmed to ISMG that they remain actively engaged with Anthropic's products, suggesting that compliance with the executive order is incomplete, inconsistently enforced, or both. Trump's directive made limited exception for the Pentagon, granting a six-month phase-out period for embedded military applications of Claude already deployed within classified defense networks, while demanding all other government use stop immediately.
The broader confrontation between the Trump administration and Anthropic centers on a $200 million Defense Department contract under which Claude had been deployed within the military's classified infrastructure. The dispute escalated sharply when Anthropic CEO Dario Amodei declined to remove safety guardrails from Claude that restricted its use in applications involving lethal force, surveillance, and sensitive data operations. Defense Secretary Pete Hegseth responded by threatening to designate Anthropic a supply chain risk — a formal classification that would effectively bar the company from future federal contracting — shortly before Trump issued the sweeping directive. The sequence of events suggests the presidential order functioned partly as a negotiating pressure instrument and partly as punitive retaliation, with Trump declaring publicly that the administration would "not do business with them again."
The persistence of Claude usage within federal agencies despite the order reveals the practical difficulties of rapidly unwinding AI tool dependencies that have become embedded in day-to-day government workflows. Unlike traditional software procurement, AI assistants are often adopted at the individual or team level with minimal centralized oversight, making blanket enforcement of cessation orders logistically complex. The absence of any publicly detailed enforcement mechanism or agency-specific compliance framework further compounds this challenge, leaving individual staffers and supervisors to interpret the directive's scope without clear institutional guidance.
The episode sits at the intersection of two accelerating tensions in AI governance: the push by national security establishments to deploy AI capabilities with minimal restriction, and the countervailing insistence by leading AI developers that safety limitations are non-negotiable features rather than contractual variables. Anthropic's refusal to strip Claude of its guardrails for military use represents one of the most direct public confrontations yet between an AI safety-oriented company and government clients demanding unrestricted operational flexibility. The outcome will set a significant precedent for how AI firms negotiate the terms of government partnerships, particularly in defense contexts where the stakes of AI-assisted decision-making are highest.
More broadly, the dispute reflects a structural ambiguity that is becoming increasingly common as AI tools proliferate across institutional settings: the gap between top-down policy directives and ground-level adoption realities. Even if the Trump administration ultimately succeeds in formally terminating its relationship with Anthropic, the continued use of Claude by federal workers underscores that AI governance is not simply a matter of executive orders or procurement decisions. Effective oversight requires operational infrastructure, monitoring capabilities, and organizational culture shifts that take considerably longer to implement than the political timelines surrounding high-profile confrontations like this one.
Read original article →