The promise of fewer officials
Ministers are looking to AI to help curb the appeal of Reform by promising to shrink the number of civil servants. At the same time, they warn that machines could exceed their bounds. That combination sounds less like a settled plan than an effort to pursue efficiency while keeping the risks at a distance.
The civil service has reason to be wary. In a Fabian Society survey, only 29% of respondents said they had been consulted about AI tools in their workplace. Before the Department for Science, Innovation and Technology (DSIT) was abolished, its permanent secretary, Emran Mian, reportedly planned to replace nearly every departing employee with AI over time rather than recruit a human successor.
At the Department for Work and Pensions, machine-learning systems are being tested to flag Universal Credit claims they consider risky. After “serious concerns” about transparency and the handling of personal data, the department says a human makes the final decision and AI does not determine eligibility for benefits.
That safeguard matters, but it does not settle the issue. Civil servants who deal directly with the public know that a human being somewhere in the process is not necessarily the person in charge.
Trust is the test
Government depends, ultimately, on public trust. Britons still trust the civil service, though trust is below the OECD average. The Public Accounts Committee has warned that poor transparency and weak standards for AI use put that trust at risk.
The fear is not primarily a sudden takeover by killer robots. It is the quiet handover of thousands of small decisions to digital officials because they are cheaper and faster, with ministers describing the change as “transformation.” The public may experience that change not as efficiency, but as a decision with no accountable person behind it.
The phrase “the computer says no,” from the television series Little Britain, still resonates partly because of the Post Office Horizon scandal. People understand the danger of a counter worker passing responsibility to a system whose decision cannot be challenged because nobody in the room knows how it was made.
That is a more immediate problem than the civil service’s age or its memory of earlier technology. The median age of civil servants is 43, and many remember offices moving from fax machines to email, then acquiring Clippy, a famously inept early attempt at a digital assistant. New tools may feel familiar; decisions made by opaque systems are a different matter.
The missing accountability
The government’s assurances are becoming harder to take at face value amid recent cases of AI going out of control. AI minister Kanishka Narayan gave assurances to Parliament’s Business and Trade Committee, while public-sector unions warn that the government is moving too quickly. The reported plan at DSIT and the trials at DWP make clear that replacing or supplementing human work is already more than an abstract prospect.
I think the central question is not whether a human signs off on a decision. It is whether that person has enough information and authority to challenge the system, and whether the citizen can get an explanation when things go wrong. “Human in the loop” describes a process; it does not tell us who is responsible.
The UK has also signalled international ambitions on AI rules. Burnham spoke at the UN about the country potentially leading work on international AI regulation, even as one of the government’s first decisions was to abolish DSIT, which had been working on AI safety. The Institute for Government warned the move would “do real harm,” while former government AI adviser Matt Clifford said it “wastes time and effort urgently needed for the work itself.”
That contrast is telling. A government can talk about international rules while letting accountability at home become harder to see. The risk is not just that AI replaces civil servants, but that citizens are left facing decisions made in their name by systems no one can clearly answer for.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X