Anthropic is deepening its foothold in the United States government even as its relationship with one part of it turns into a full-blown legal battle. On Tuesday, the company and the General Services Administration announced an expanded OneGov agreement that gives federal agencies a 40% discount on Claude Enterprise, Claude for Government, and the Claude API — while the company's fight with the Pentagon over its "supply chain risk" designation shows no sign of cooling.
The agreement, detailed in an Anthropic blog post published October 27, applies the discount across every product surface for the first time, including purchases made directly from Anthropic and, in the coming months, through cloud providers. It also extends beyond Washington: state, local, tribal, and territorial governments are eligible for the same terms. For more coverage of how governments are shaping the AI market, follow our latest AI developments.
What the new deal changes
The previous OneGov arrangement offered agencies Claude for Government at $1 per user, an offer Anthropic had already extended to the end of October. The new agreement replaces that seat-based model with consumption-based pricing — agencies pay for the tokens they use, with no minimum spend commitment. GSA said the discounts are estimated to save the government $1 billion.
The shift mirrors the private sector, where consumption-based pricing has largely displaced per-seat licensing. It also follows OpenAI, which moved to consumption-based pricing as part of its own OneGov deal last month, according to FedScoop. Google's Gemini government offer — currently priced at $0.47 per agency — is set to expire November 15, and GSA has not yet detailed a renewal.
Not everyone is convinced the model is a win for taxpayers. Analyst firms including Gartner and Forrester have warned that token-based pricing makes costs hard to predict, since vendors vary in how they calculate and bill consumption. "A small group of power users can drive a disproportionate share of costs, turning successful AI adoption into both a productivity win and a budgeting challenge," Tracy Woo, principal analyst at Forrester, wrote in an August blog post cited by FedScoop.
Adoption is growing fast
By Anthropic's own accounting, the government business is scaling quickly. Roughly 500 US federal, state, and local government organizations now use Claude. Weekly active users of Claude for Government are up about eightfold since February, and usage across federal civilian agencies has grown roughly twentyfold in a year, the company said.
Anthropic also pointed to concrete deployments: Utah gave Claude Code to 34 developers across 10 state agencies, with 71% reporting savings of 20 or more hours by the end of a four-week pilot. Anthropic says a state CIO reported responding to a ransomware attack in eight minutes rather than the usual two to three days, and that a federal security team used Claude to detect and patch a vulnerability — one a human-led penetration test had missed — in about two days.
The Pentagon fight keeps escalating
The civilian expansion stands in sharp contrast to the company's defense-sector troubles. The dispute began after Anthropic resisted allowing its technology to support fully autonomous weapons and mass domestic surveillance, while Pentagon officials demanded acceptance of all lawful uses, according to Military.com's account of the court record.
In September, the US Court of Appeals for the DC Circuit ruled 2-1 that the Pentagon reasonably excluded Claude from its supply chain, accepting concerns that company-imposed restrictions could prevent the technology from performing military tasks, and rejecting Anthropic's argument that the designation retaliated against its advocacy for AI regulation. A separate California case went the other way in August, when a federal judge struck down a parallel designation under a different statute.
Anthropic sought a rehearing on October 13, arguing that the court had stretched a law aimed at hostile foreign actors into a tool that could be used against any American contractor that disagrees with the Pentagon. The tension is compounded by an October 9 disclosure that Anthropic models took unintended actions on government websites during testing — including submitting an invented homicide tip — a reminder that both over-restriction and unintended action carry real costs.
The political temperature keeps rising alongside the legal one. ABC News reported Wednesday that President Trump ordered the US government to cut ties with Anthropic, with Defense Secretary Pete Hegseth declaring the company a supply chain risk, while Bloomberg described Anthropic's tensions with the administration as laid bare ahead of its landmark IPO. The company has, for now, a split-screen relationship with the federal government: welcome at GSA, HHS, and state capitols, and increasingly unwelcome at the Pentagon.
Why the timing matters
All of this is unfolding in the run-up to what is expected to be one of the largest technology IPOs in history. Government contracts are a material part of the Anthropic story investors are being asked to buy — and a material risk, too. An expanded civilian deal with an estimated $1 billion in savings strengthens the revenue narrative; a spreading blacklist fight raises the question of how much of the public-sector market the company can actually keep.
For GSA, the agreement is part of a broader push to lock in discounted AI access for agencies before promotional deals expire. "As offers expire, GSA will renew agreements where appropriate and work to ensure continued access to critical tools while delivering the best value for taxpayers," the agency said in a statement.
For Anthropic, the calculation is more delicate. Every new agency that builds on Claude makes the company harder to dislodge from civilian government — and gives it more leverage in the fight over where, and under what conditions, its models may be used in defense.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
