Artificial intelligence (AI) is steadily embedding itself across local government. From triaging housing applications to supporting social care assessments and analysing planning data, AI promises greater efficiency at a time when resources remain under acute pressure.

Yet, for all its potential, AI does not sit comfortably within the traditional frameworks of public law. The legal duties governing decision-making have not changed, but the means by which decisions are reached are evolving in ways that make compliance harder to evidence, and failure more difficult to detect.

For local authorities, the question is no longer whether AI will be used, but whether it can be used in a way that is consistent with the transparency, accountability and fairness expected of public bodies.

The enduring importance of judgment

At the heart of public law lies a simple proposition: decisions must be taken by the decision-maker, lawfully exercising their discretion. AI complicates this in subtle ways.

In practice, many systems generate outputs that appear authoritative, for example risk scores, prioritisation rankings, or eligibility indicators. Where these outputs are persuasive, or where workloads are high, there is a natural tendency for decision-makers to defer to them. The legal risk does not arise from the use of the tool itself, but from the extent to which human judgment is displaced.

Courts have long been wary of rigid decision-making structures. The introduction of AI risks recreating those concerns in a modern form. Decision-makers may not consciously fetter their discretion but may do so inadvertently by treating algorithmic outputs as determinative. That sits squarely in the territory of fettering discretion discussed in British Oxygen Co Ltd v Minister of Technology [1971] AC 610 and R v Port of London Authority, ex p Kynoch Ltd [1919] 1 KB 176. The point is not that policies and tools are unlawful, but that discretion must remain real, and exceptions must genuinely be considered.

The challenge for authorities is evidential as much as substantive. It will rarely be enough to assert that a system is 'advisory'. The real question is how decisions are made in practice, and whether the workflow encourages meaningful scrutiny or passive acceptance. Where outcomes look irrational in the legal sense, the familiar principles of rationality review, commonly traced back to Associated Provincial Picture Houses Ltd v Wednesbury Corporation [1948] 1 KB 223, remain the yardstick, even if the reasoning chain runs through a model.

Equality in an automated environment

The Public Sector Equality Duty (Equality Act 2010, s 149) requires decision-makers to have due regard to the impact of their decisions on protected groups. That duty is inherently forward-looking and context-specific. AI systems, by contrast, are typically trained on historical data.

This creates a tension. Where underlying data reflects entrenched inequalities, whether in housing allocation, enforcement activity or service access, there is a risk that those patterns are reproduced, or even amplified, through automated processes.

The difficulty is not simply identifying bias but recognising it in time. Unlike conventional policies, where impacts can be assessed through consultation and analysis, algorithmic systems may produce outcomes that are not immediately transparent, particularly where their internal logic is opaque.

Case law on the equality duty is helpful here because it focuses on decision quality rather than good intentions. The 'Brown principles' from R (Brown) v Secretary of State for Work and Pensions [2008] EWHC 3158 (Admin), and the Court of Appeal’s emphasis on rigorous, properly informed engagement in R (Bracking) v Secretary of State for Work and Pensions [2013] EWCA Civ 1345, are readily engaged by AI deployment. If a system is adopted without genuinely understanding likely impacts, or if equality analysis is treated as a procurement tick-box, the authority is exposed.

The most direct illustration of the courts treating automated tools as a live equality and governance issue is R (Bridges) v Chief Constable of South Wales Police [2020] EWCA Civ 1058. While not a local authority decision-making case, it is a clear reminder that public bodies are expected to evidence their thinking, including on equality impacts, when using advanced technology in operational settings.

Authorities therefore face an ongoing obligation: not only to assess equality impacts at the point of implementation, but to monitor outputs continuously and intervene where necessary. Equality analysis becomes less of a one-off exercise and more of a continuing obligation.

The problem of explanation

Transparency sits at the centre of lawful decision-making. Individuals affected by decisions are entitled to understand, at least in broad terms, how those decisions were reached. This is essential both to procedural fairness and to the ability to challenge decisions where appropriate.

AI presents an obvious difficulty. Some systems, particularly those based on complex machine learning techniques, cannot readily produce clear, human-readable explanations. Even where explanations are available, they may not align neatly with the legal requirement to identify the key factors underpinning a decision.

For local authorities, this creates a practical dilemma. Reliance on systems that cannot be adequately explained risks undermining the ability to give lawful reasons. Over-simplification, on the other hand, risks mischaracterising the role the system played.

In this context, explainability is not merely a technical issue; it is a legal one. In planning, the standards for adequate reasons are often discussed through South Bucks District Council v Porter (No 2) [2004] UKHL 33. More broadly, fairness can require reasons depending on context, as reflected in R v Secretary of State for the Home Department, ex p Doody [1994] 1 AC 531. AI does not replace these principles. It makes them harder to meet unless the authority designs the process around them.

There is also a parallel set of requirements under data protection law. Where personal data is processed, the UK GDPR transparency duties (Articles 13 to 15) and the constraints around solely automated decisions with legal or similarly significant effects (Article 22) are the obvious pressure points. Even where Article 22 is not engaged because a human is formally involved, authorities still need to be able to explain what the tool did, what data it used, and what role the output played in the final decision.

Authorities must be confident that they can articulate the basis of decisions clearly enough to satisfy a complainant, an auditor, the Ombudsman, or a judge.

Data, discretion and control

AI systems depend on data, often large volumes of it, drawn from multiple sources. That reliance raises familiar data protection issues, but also more subtle questions about control and accountability.

Where data is shared across departments, or processed using third-party platforms, the boundaries of responsibility can become blurred. Decisions may be shaped by inputs that are neither fully understood nor easily interrogated by the authority itself. That is risky in ordinary public law terms too. Decision-makers are under a duty to take reasonable steps to inform themselves before deciding, a principle associated with Secretary of State for Education and Science v Tameside Metropolitan Borough Council [1977] AC 1014. If an authority cannot explain the provenance, relevance, or quality of key inputs, it may struggle to show the decision was properly informed.

At the same time, the accessibility of AI tools creates the risk of informal or unregulated use. Officers may turn to widely available systems to assist with analysis or drafting, without any clear governance framework in place. The legal implications of such use, particularly where personal data, special category data, or confidential information is involved, can be significant. Data protection compliance also becomes harder to evidence without a clear record of what was used, by whom, for what purpose, and with what safeguards. In addition to UK GDPR obligations, the Data Protection Act 2018 contains the domestic scaffolding for risk assessment and governance, including provisions tied to DPIA practice.

What emerges is a broader governance challenge: ensuring that AI use is not only compliant in principle, but also visible, controlled and auditable in practice.

A shift in risk, not in principle

None of this suggests that AI is incompatible with public law. On the contrary, many of the risks identified are familiar, albeit in a different guise. Issues of consistency, fairness and transparency have always been central to local authority decision-making.

What AI changes is the scale and subtlety of those risks. Decisions may be taken more quickly, on the basis of more complex information, and with less direct human engagement with the underlying reasoning. Errors, when they occur, may be harder to detect and more widely replicated.

This places a premium on governance. Policies, training and oversight mechanisms, often treated as supporting functions, become central to ensuring that legal duties are met.

Looking ahead

The direction of travel is clear. AI will play an increasing role in how local authorities deliver services and make decisions. Financial pressures and rising demand make that trajectory difficult to resist.

The legal framework, however, remains rooted in principles that demand accountability, transparency and the exercise of independent judgment. Those principles are technology-neutral, but not technology-proof.

Authorities that approach AI as a purely technical solution risk underestimating its legal implications. Those that treat it as a governance challenge, embedding legal, equality, data protection and operational considerations from the outset, are more likely to realise its benefits without compromising their public law obligations.

There is a further dimension. The common law of England and Wales is inherently adaptive, with principles tested and refined through individual cases. That makes it well suited to absorbing the challenges AI presents, but it also means that AI, trained on existing law, cannot reliably anticipate how the law will develop. We can expect the courts to address these questions incrementally, not by restating fundamental principles but by providing fact-specific guidance on the procedural and governance safeguards required to make the use of AI defensible within public law decision-making. In that sense, the law in this space may well adapt before our eyes.

In short, AI does not lower the standard against which decisions are judged. It simply offers a more expensive way of failing to meet it.

Browne Jacobson has been both using AI when appropriate and advising on the legal, governance and data protection dimensions of AI adoption for some time. If your authority is considering how best to deploy AI tools or would benefit from training on the issues discussed in this article, we would be happy to help.