Trump banned Anthropic from the entire federal government. Now both sides want a deal.

On February 27, 2026, the Trump administration told every federal agency to immediately stop using Anthropic technology, a blanket ban affecting one of the most widely deployed AI systems in the US government. Less than four months later, White House officials are sitting across the table from Anthropic CEO Dario Amodei, trying to figure out how to reverse it.

The two sides met on April 17, 2026, in what officials described as a "productive" conversation covering Anthropic's AI models, safety protocols, and federal usage terms. That meeting wasn't a one-off: additional sessions with Commerce Department officials were scheduled for June 15-16, 2026, suggesting a sustained negotiation, not a photo opportunity. The agenda reportedly covers three areas: safety protocols that satisfy national security concerns, international access frameworks for Anthropic's models, and terms under which federal agencies might resume using the company's technology.

To understand why that last item matters, you need the full chain of events. The dispute began when the Defense Department demanded unrestricted access to Anthropic's AI for military applications, including uses the company believed could enable autonomous lethal weapons systems and mass domestic surveillance.

Anthropic refused. The Pentagon responded by formally designating Anthropic a "supply chain risk", a label that doesn't just cut off one contract. It effectively signals to every federal procurement office, contractor, and supplier that doing business with the company carries political and legal risk. Then came Trump's February 27 order, which extended the ban government-wide and included a six-month phase-out period for agencies like the Defense Department that had already integrated Anthropic's Claude models deeply into their operations.

By mid-June, the administration escalated further. New directives restricted foreign access to two specific Anthropic models, Fable 5 and Mythos 5, citing jailbreak risks: the possibility that adversaries could manipulate the models' safety mechanisms to extract dangerous capabilities. The restrictions effectively made Anthropic's most advanced models unavailable to international partners and allies who had been using them, adding a geopolitical dimension to what had been a domestic regulatory fight.

Anthropic's position throughout has been that its guardrails aren't a commercial preference, they're the core of what the company is. Founded in 2021 by former OpenAI researchers including Amodei and his sister Daniela, Anthropic built its entire brand around the premise that powerful AI models need built-in safety constraints. Stripping those constraints to satisfy Pentagon demands would, from Anthropic's perspective, undermine the thing that makes its models worth using. The company didn't back down when Trump ordered the ban. That it's now in negotiations at all signals that the administration, not just Anthropic, wants a resolution.

The political pressure helping move things forward came from an unexpected place. Senate Armed Services Committee Chairman Roger Wicker and ranking Democrat Jack Reed, along with Senate Defense Appropriations leaders Mitch McConnell and Chris Coons, sent a private letter to both Anthropic and the Pentagon urging them to resolve the dispute, a rare bipartisan intervention in an AI policy fight. When senior Republican senators start pushing the administration to make peace with an AI safety company it just banned, the calculus has shifted.

What the talks produce, if anything, will set a precedent for how the US government handles AI companies that refuse to remove their safety guardrails on demand. Every major AI lab is watching. OpenAI, Google DeepMind, and Meta all have active government contracts, and all of them have their own versions of the same underlying tension: how much control do you hand a government client before you've compromised the product itself? Anthropic drew a line. The question now is whether Washington decides to respect it.