The scale of the federal government's AI adoption push is only now becoming clear, and it is larger than most outside observers had appreciated. ChatGPT is being made available to the entire US federal workforce. Anthropic has offered Claude — at one dollar — to all three branches of government, including the legislative and judicial branches. OpenAI just expanded its government footprint through a new deal with AWS, announced this week. The White House budget office has directed every corner of the executive branch to deploy AI, declaring that the federal government will "no longer impose unnecessary bureaucratic restrictions" on AI use. The speed and breadth of this transition is unusual by any historical standard for government technology adoption.
The Washington Post's inside look at the Trump administration's AI push, published last month, found a federal bureaucracy in the early stages of a genuine transformation — one driven more by directive than by organic adoption. Agencies that had spent years running cautious AI pilots found themselves being pushed to deploy at scale, quickly. The results have been mixed in ways that are predictable for anyone who has watched enterprise AI adoption: places where the tools fit naturally into existing workflows have seen real productivity gains, while places where they were deployed without adequate training or process design have seen frustration and workarounds.
Pennsylvania's experience with ChatGPT offers a useful ground-level data point: state employees participating in the pilot found it reduced time on routine tasks by about 105 minutes per day. Extrapolated across a large bureaucracy, that represents a significant labour efficiency gain. The question is whether routine task reduction translates into better public services, or simply into the same services delivered with fewer people. That question — whether AI in government makes services better or just cheaper to provide — is not one the current administration has been keen to dwell on publicly.
Anthropic's offer of Claude to all three branches of government at a dollar is notable both for the price and for the scope. The legislative and judicial branches have historically been slower to adopt new technology than the executive, partly because their IT infrastructure is older and more fragmented, and partly because the nature of their work — lawmaking and adjudication — involves more judgment and less routine processing than agency administration. Anthropic has certified Claude at FedRAMP High, the most stringent requirement for handling unclassified sensitive government data, which removes a significant procurement obstacle. At a dollar, the barrier to experimenting is essentially zero.
OpenAI's AWS deal, reported on Tuesday, is part of a broader pattern of AI companies building the infrastructure layer for government deployment. Amazon has committed up to $50 billion in AI services for the US government, with data centre construction breaking ground this year. The logic is straightforward: government is a large, stable customer with predictable long-term spending, and the contracts tend to be sticky. Getting embedded in federal procurement at this stage of the AI adoption curve creates durable relationships that are difficult for later competitors to dislodge.
What is less visible in the press releases is the question of accountability. When a federal employee uses ChatGPT to draft a benefit determination, or a judicial clerk uses Claude to summarise case law, who is responsible for errors? Government AI deployments raise oversight and audit questions that private-sector deployments largely don't — decisions by government agencies affect people's legal rights in ways that private-sector AI decisions generally don't. The speed of the current adoption push has outpaced the governance frameworks that would normally accompany it, and that gap is a live risk that the enthusiasm around productivity gains has tended to obscure.