The UK government has selected Anthropic to build an artificial intelligence assistant that will handle administrative tasks for citizens, from job applications to GP registrations, in what officials describe as the first national-scale deployment of agentic AI technology.
The Department for Science, Innovation and Technology signed a memorandum of understanding with the San Francisco-based company in February to develop a pilot program using Claude, Anthropic’s flagship model. Unlike chatbots that simply answer questions, the system will actively complete tasks—filling forms, booking appointments, and routing users to appropriate government services.
The initial pilot focuses on employment services, helping citizens find work, access training programs, and navigate benefits. But the ambition extends far beyond job searches. Officials envision the assistant handling life transitions like moving homes, registering to vote, or switching doctors—moments when bureaucratic friction often leaves people frustrated or confused.
“Government aims to rethink and reshape how public services help people through crucial life moments using the power of emerging AI technology,” Technology Secretary Peter Kyle said.
The project follows the government’s “Scan, Pilot, Scale” framework, with potential nationwide rollout targeted for late 2027. It forms part of the AI Exemplars Programme, which projects £45 billion in productivity gains from artificial intelligence adoption across public services.
What distinguishes this initiative is the technology’s active role. Government documents note the system is “unique in that it can reliably complete basic admin tasks for people as well as provide tailored support.” The assistant retains context across sessions, learning user circumstances to provide personalized career guidance and service recommendations.
The UK AI Safety Institute has conducted active testing of Claude 3.5 Sonnet, the model powering the assistant, in joint assessment with US counterparts. Evaluators identified some jailbreak vulnerabilities during testing, though the system proceeded to deployment with safety protocols in place. Users will maintain data control and opt-out options, while phased rollout includes evaluation checkpoints before expansion.
The government has structured the partnership to avoid vendor lock-in. Anthropic engineers work alongside civil servants in a knowledge-transfer arrangement, with the state maintaining product ownership and building internal AI capabilities. Industry observers note this approach matters more than the underlying technology itself.
“Success is less about the underlying model and more about the governance, data architecture, and internal capability built around it,” according to analysis from AI News.
The UK move positions the country as a first-mover in government AI agents, though Anthropic has been building a public-sector portfolio globally. The company secured a $200 million agreement with the Department of Defense and negotiated $1 access to Claude for all US government branches. Similar pilots are underway in Iceland and Rwanda, while Lawrence Livermore National Laboratory reports 10,000 scientists using the system.
The deployment tests whether AI can bridge the gap between information availability and user action—a persistent challenge in digital government services. Citizens often know services exist but struggle to access them due to complexity, scattered information, or administrative burden.
Whether the assistant succeeds in moving people from awareness to completion will determine if other governments follow Britain’s lead. The 2027 target date provides a tight timeline to demonstrate that agentic AI can operate reliably at population scale while maintaining safety standards and public trust.