Why AI matters across public services now

AI has moved from pilot projects to government policy. In January 2025 the government set a target of £45 billion a year in productivity savings from digital technology and AI across the public sector. A cross-government trial of AI assistants involving 20,000 civil servants reported average self-reported savings of nearly two weeks a year per person. Similar programmes are under way in the NHS, local government and the emergency services. Policing, for example, now has a national AI centre, PoliceAI, set up to pilot and scale tools across every force in England and Wales.

Government has also set out the standard it expects. The AI Playbook for the UK Government, published in February 2025, sets ten principles for safe and responsible AI use in public sector organisations. They include knowing what AI is and what its limits are, using it lawfully, ethically and responsibly, and having meaningful human control at the right stages. Central government departments and many arm’s-length bodies must now publish details of the algorithmic tools they use under the Algorithmic Transparency Recording Standard, and other public bodies are encouraged to do the same. Sector bodies have added their own commitments, such as the National Police Chiefs’ Council’s Covenant for Using AI in Policing.

All of this rests on long-standing legal duties. Public bodies must comply with data protection law, including the rules on automated decision-making as amended by the Data (Use and Access) Act 2025. They must also meet the public sector equality duty under the Equality Act 2010 and act fairly and rationally in public law. Wherever public money and public power meet AI, leaders are accountable for what the technology does in their name.

Why leadership and governance decide the outcome

Public services depend on legitimacy: the public’s belief that public power is used fairly, lawfully and for the right reasons. AI does not change that test. It raises the stakes. An unchecked tool can repeat an error quickly and at scale, and the public will hold the organisation to account, not the supplier.

The risks are not hypothetical. In 2020, A-level grades calculated by a statistical model were withdrawn within days after public outcry, and students were given their teachers’ assessed grades instead. In 2024, a fairness analysis by the Department for Work and Pensions of a machine learning tool used to flag Universal Credit claims for review found statistically significant disparities linked to age, disability, marital status and nationality. In late 2025, a police intelligence report supporting a ban on away supporters at a Birmingham football match cited a fixture that had never taken place. The detail had been generated by Microsoft Copilot, and the force’s chief constable retired after the error became public.

What these cases share is a gap in governance, not just a faulty tool. Outputs were not checked rigorously enough, safeguards were tested too late, or leaders could not fully account for what the system was doing. The courts have drawn the same line. In R (Bridges) v Chief Constable of South Wales Police (2020), the Court of Appeal found a police force’s use of live facial recognition unlawful because it left too much discretion to individual officers, relied on an inadequate data protection impact assessment, and had not met the public sector equality duty. The principle applies to any public body: technology that may be useful still has to be governed.

Good AI leadership means being able to answer a few plain questions before a tool goes live, and keep answering them afterwards:

• Who is accountable for each decision the tool informs, and can they explain it to a court, a tribunal, an ombudsman, an auditor or a member of the public?
• Where does a human stay in the loop, and do they have the time and skill to challenge the output?
• Has the tool been tested for bias, and has the equality impact been assessed?
• Have the staff who will use it been consulted and trained, and have the people affected been told?
• What will trigger a pause, a review or a switch-off?

These are leadership questions, not technical ones. They are exactly what the CMI AI qualifications teach managers at every level to ask, and to act on.