White House preps order to boot Claude from federal agencies

If you are working on federal projects using Claude, you might need to find a new assistant very soon.

The White House is preparing an executive order formally instructing the federal government to rip out Anthropic's AI from all operations, Axios reported. A source familiar with the matter said the order could land as soon as this week.

Some parts of the government aren't even waiting for the signature. Agencies like the Treasury Department have already started offboarding Claude.

Safety guardrails meet military pushback

The move marks a steep escalation in the administration's clash with the AI lab. President Trump previously declared that his administration would not use "woke" AI, leading the Pentagon to slap Anthropic with a supply chain risk designation.

Anthropic didn't take that sitting down. On Monday, the company sued the Pentagon, arguing that federal procurement laws give the administration no authority to blacklist a domestic company over protected speech.

White House officials see it differently. The administration argues that Anthropic's built-in safeguards pose a genuine national security threat, specifically worrying that commercial AI guardrails could allow tech companies to intervene during active military operations.

A White House official downplayed the leaks, saying any policy announcement will come directly from the president and that talk of an executive order remains speculation.

An unprecedented move against a home-grown player

Using executive orders to target tech is not new for Trump. In his first term, he leaned heavily on national security claims to restrict foreign firms like TikTok and Chinese telecom giants.

Targeting a prominent U.S. company by name, however, is almost entirely unprecedented outside standard procurement channels. Even with Huawei, Trump avoided naming the vendor outright in his original order, leaving that step to Congress.

Why it matters

For federal workers, contractors, and developers relying on Claude's industry-leading coding and reasoning benchmarks, this forces a messy migration. Ripping out deployed software pipelines is neither cheap nor quick.

For the broader AI ecosystem, the implications look chilling. If built-in safety boundaries can be redefined as national security risks whenever politics get contentious, every AI lab bidding on public contracts will feel the squeeze.

We think this approach sets a messy precedent. Treating model guardrails as corporate defiance during military operations seems like an aggressive stretch to sideline a vendor over ideological friction. If presidential power can bypass established procurement law to ban domestic tech innovators, it risks teaching the entire industry to ditch safety research just to keep government contracts.