AI in local government: Key legal insights from recent training

Browne Jacobson partners Anja Beriro, James Arrowsmith and Richard Nicholas recently delivered a two-part training programme on AI adoption in local government to EM LawShare members.

The sessions brought together local government lawyers to explore how legal teams can support their organisations in navigating the strategic, governance and practical challenges of implementing AI.

Why this matters now

Responses gathered in connection with the training showed many organisations are at the beginning of their AI journey, with aims to develop capacity or have some good practice within two years. Strategy and governance emerged as key barriers, with work still to be done on compliance areas such as secure by design, data governance and appropriate contracts. There was also concern about AI outpacing governance and compliance, with associated regulatory risk.

The message from the training was clear: whilst AI offers significant potential for efficiency and innovation, organisations need robust frameworks in place before they can safely harness these benefits.

What the polls revealed

Live polling during the sessions painted an interesting picture of where local government currently stands. 

Most delegates indicated their organisations are using AI in some form, though there was variation in whether legal teams themselves are using AI tools and how involved they are in organisational AI adoption decisions.

Objectives are strongly geared towards efficiency, productivity and cost saving, with innovation also getting some interest. Uses often include AI assistants, magic notes and similar tools, with primary barriers being lack of strategy, capacity and capability.

When asked about documented outcomes and priority use cases, most respondents said they had "some pilots only" or "not yet", with very few having established use cases across multiple services. Similarly, on governance infrastructure (AI strategy, principles, governance board and training plan), most reported limited progress to date.

Stronger domains included contract and risk, challenges and disputes, and business case and contract management, whilst more challenging areas were strategy, governance, procurement, and implementation and piloting.

The core framework

Rather than thinking about AI as a linear journey, we recommended organisations think in terms of a comprehensive framework that can be applied flexibly depending on where you're being asked to provide support. The framework provides a structure to approach AI portfolio management, from vision, through to governance and compliance, to stakeholders, maintenance and exit strategies. 

The key insight? While AI does have many things in common with conventional IT implementations, the different behaviour of AI tools do require a change in the way we approach legal aspects of implementation. These include the way AI evaluates information rather than applying rules to determine outcomes, and the manner in which the data, technology and uses can all evolve over time. The framework we explored offers a structure to manage this well.   

Practical recommendations

Start with vision and strategy

Decision makers and staff need to understand the destination and the roadmap. This means deciding whether you're pursuing general AI or a portfolio of use cases, a secure perimeter or rules-based approach, and innovation in teams or centrally led. Your strategy should support delivery of this vision.

Build governance that works

Effective governance encompasses planning, communications, governance structure, training, risk assessment and controls. A consistent compliance framework increases certainty for staff in relation to tools, whilst a regulatory oversight process (with monitoring) helps them understand their role in supporting compliance. 

Get procurement and contracts right

AI solutions learn and evolve over time, with each generation being unique and outputs hard to predict in advance. Suppliers may want the right to access results of the evolution. This means you need a responsible person to monitor, ongoing testing through 'red teaming', and a clear understanding of what you're comfortable with regarding provider access to personal data.

Understanding the difference between the 'AI model' (the engine) and the 'AI system' (the car, which includes system prompts, RAG and training) is crucial. Suppliers may not be able to change the model but should be responsible for parts of the system.

Focus on workforce alignment

This requires literacy work to support understanding of tools and guidance, development of an organisation-wide AI use policy, and establishing training and an AI hub.

Don't forget communications

Monitor new features and communicate your organisation's position on those as soon as possible to avoid unauthorised usage. Plan training sessions to upskill staff on using and maintaining AI components when they are approved for firm-wide use.

Looking ahead

The regulatory landscape is evolving rapidly, and AI implementations in the public sector receive close scrutiny. We are going to see very public stories of AI risks and failings, while AI success will not get the same profile. Public bodies need AI to deliver for the public, but need to be confident that AI is deployed in safe and legal ways. 

AI is also beginning to drive scrutiny. The public have access to AI tools to challenge planning decisions. AI drafted complaints are on the increase, and regulators are invest in AI to support their oversight functions. Robust and defensible decision making has never been more important. 

The bottom line? Successful AI adoption requires legal teams to move beyond reactive advice to strategic partnership, helping organisations build the governance, compliance and risk management frameworks necessary to harness AI's potential whilst protecting public trust.

Continue the conversation

Richard Nicholas hosts AI legal bytes, a bi-weekly half-hour discussion on key AI topics for lawyers. It's a great way to stay current and connect with peers navigating similar challenges.