TL;DR
A Global Government Forum study built on interviews with 13 permanent secretaries and senior AI leads, launched in London on 15 October at Public Service Data.AI, a conference GGF co-hosts with the government, argues that Whitehall has left the “foothills” of AI and now needs an operating model rather than more encouragement to adopt. It names four priorities: setting ambition and measuring impact, joining up to scale what works, reshaping the workforce, and setting the parameters that let trust and pace coexist. It arrives a day after the Public Accounts Committee chair compared AI’s approach to government with a tsunami.
The report, “The new AI operating model for government”, had the former cabinet secretary Lord Gus O’Donnell as its lead, and follows GGF’s 2025 “Rewiring the State” work on barriers to digital transformation. Interviewees include Cat Little, John-Paul Marks and chief AI officer Kalbir Sohi. The study’s knowledge partners were Amazon Web Services and Cognizant, and their joint foreword from the two companies calls for AI to be built into “secure cloud-based infrastructure” rather than left in pilots.
The four priorities
On ambition, the study finds that broad adoption will bring efficiency gains but that deeper change to services will not happen on its own; leaders must choose where AI can make the biggest difference and find ways to measure it. One interviewee framed the task as “establishing the tooling to scale AI rather than just playing with it”.
On coordination, the report says leaders often cannot see what other departments are doing while each buys tools for needs they share, and it argues for more central direction where common approaches would cut duplication, with flexibility where needs genuinely differ.
On the workforce, it asks leaders to look beyond skills training to how roles, career paths and whole professions will change, and what that means for recruitment. On trust, interviewees wanted clarity on risk appetite and accountability, and several warned that the public will forgive an AI error less readily than a human one.
The wider picture
The study lands in a crowded week. We reported the PAC chair’s warning about legacy IT, Scottish civil servants’ complaints about AI-drafted questions, and today the Home Office’s spot checks on its asylum summariser. GGF notes that Antonia Romeo, the cabinet secretary, has called AI “the defining opportunity of our generation, as well as the defining challenge”, and that Andy Burnham has set up an AI taskforce and put the AI minister in Cabinet.
Looking forward
In our view the report’s strongest point is also its least comfortable: the centre-versus-department tension it describes is exactly what the asylum and Scottish cases illustrate, where individual departments moved fast and the quality assurance did not follow. Readers should weigh the sponsorship: a cloud provider and an outsourcer underwriting a call for scaled, cloud-based AI are not disinterested. O’Donnell’s foreword concedes that gains from new technology take time and need redesigned workflows and retraining, the candid version of the pitch.