The U.K. government is preparing to put artificial intelligence deeper into the machinery of the public sector. A week after announcing a broader plan for major AI investment, it is setting out how that ambition could affect civil servants, public services and the way departments share data.
The centerpiece is Humphrey, a new set of AI tools named after a character from the U.K. political sitcom Yes, Minister and Yes, Prime Minister. The goal is straightforward: reduce bureaucracy, speed up government work and modernize systems that officials say are still wasting time and money.
A public-sector AI push takes shape
The plans are due to be formally unveiled at a press conference Tuesday led by the Department of Science, Innovation and Technology, alongside Work and Pensions and Health/Social Care. The projects appear to be at an early stage, with some work still framed as a commitment rather than a fully deployed service.
For example, the plan to bring more AI services into the customer-facing side of the NHS is described as being at the stage of a “charter” committing to the concept. Other parts of the program already point to GitHub repositories where some work can be reviewed.
Several important implementation details remain unclear. The source article says it is not yet known how many people are working across the projects, or which third-party tools, including LLMs, are being used.
What is clear is the government’s stated motivation. DSIT says the government currently spends some £23 billion annually on technology. The department’s argument is that this money should be redeployed in a more modern way, with AI and automation used to remove friction from public administration.
“Sluggish technology has hampered our public services for too long, and it’s costing us all a fortune in time and money… Not to mention the headaches and stresses we’re left with after being put on hold or forced to take a trip to fill out a form,” said Peter Kyle, the Secretary of State for DSIT, in a statement. “My Department will put AI to work… We will use technology to bear down hard to the nonsensical approach the public sector takes to sharing information and working together to help the people it serves.”
What Humphrey is meant to do
Humphrey is not described as a single chatbot. It is a set of applications aimed at helping government employees handle the heavy reading, summarizing, note-taking and document preparation that can dominate civil service work.
The tools named in the plan each target a specific task:
- Consult is designed to read and summarize “thousands” of consultation responses in hours.
- Parlex will let government employees query and read conversations in Parliament relevant to bills or other policy documents they are working on.
- Minute is a secure transcription service for meeting notes.
- Redbox helps prepare briefings and policy documents.
- Lex is focused on helping government workers find relevant legal data.
The common thread is information overload. Consultations can generate long and numerous responses. Parliamentary conversations can be difficult to search at speed. Briefings and policy documents often require civil servants to pull together facts, context and prior material from large bodies of information.
If the tools work as intended, they could make routine government knowledge work faster. The larger question is how much weight officials will place on the outputs, especially when the source article notes that some early apps are in testing phases only.
Public services are another target
The AI plan is also aimed at services the public deals with directly. DSIT is focusing on legacy bureaucracy in areas where people still spend time on calls, visits and paperwork-heavy processes.
The source article gives several examples: the 100,000 calls that tax authorities get daily, the need for people to appear in person to register a death, and the requirement to post ads in local papers as part of getting a license to drive a truck.
DSIT’s view is that overhauling processes like these with more AI-fueled automation could save £45 billion annually. The article notes that it is not clear whether that estimate is before or after the cost of building and running the AI services.
That distinction matters. A service can look efficient on paper while still requiring major investment, maintenance, oversight and support. The plan’s impact will depend not only on whether AI can automate a task, but whether the redesigned service actually works better for the people using it.
Data sharing could be powerful and risky
The third part of the push is collaboration between departments. The government wants data to move more easily across public bodies so services can be procured and delivered faster.
DSIT describes the principle as “a common-sense approach to sharing information.” One example in the source article is that central government departments, such as HMRC and the Department for Business and Trade, could share data with each other and local councils in fraud investigations.
That kind of data sharing could make government work faster in some cases. It also raises a serious question: what happens to data protection for individuals when information is shared in unintended ways?
This is one of the central tensions in the plan. The same data movement that may help departments work together can also increase the risk that personal information travels beyond the context in which people expected it to be used.
The hard part is execution
The AI program signals that the U.K. government wants to move beyond broad ambition and into public-sector deployment. It is creating a new team within DSIT to head the projects, described in the source article as a little like DOGE in the U.S., but conceived of and run by government people rather than tech moguls.
Still, the plan faces more than technical questions. The source article points to the human and institutional challenges of making programs work across departments. A former civil servant who now works for an AI company notes that past efforts to create programs spanning departments have not always succeeded.
The levers are practical: collaboration, money and authority. AI tools may summarize documents, transcribe meetings and surface legal data, but the broader program will depend on whether departments can coordinate around shared goals and trust new systems without losing accountability.
For now, Humphrey is best understood as an early test of how far the U.K. government wants to bring AI into everyday administration. The promise is less waiting, less duplicated work and faster access to information. The risks are equally clear: weak data protection, overreliance on AI conclusions and the familiar difficulty of changing how government departments work together.