Summary
- The government's Local AI team interviewed officials from more than 15 councils about developing and implementing AI policies.
- Devon, Denbighshire, and Leeds illustrate different approaches to oversight, tool approval, staff guidance, and training.
- A shared AI policy template and Responsible AI Charter are being developed with councils, with publication anticipated by spring 2027.
Local authorities are introducing approval procedures, oversight boards, and technical restrictions to manage staff use of artificial intelligence, as new government research identifies how councils are putting AI policies into everyday practice.
The Ministry of Housing, Communities and Local Government has examined the experience of officials from more than 15 councils who were involved in developing or implementing policies governing AI. The interviews covered organisations preparing their first policy and others already revising established arrangements as technology and staff requirements changed.
Published on 12 November, the findings describe approaches to software approval, ethical oversight, staff training, and public transparency. They will inform a shared AI policy template and a Responsible AI Charter being developed by the government’s Local AI team in collaboration with local authorities.
The proposed resources are expected by spring 2027, although that remains an anticipated publication window rather than a binding implementation deadline for councils.
Approval processes put policies into practice
Denbighshire County Council has connected its AI policy to the process employees use when requesting access to AI tools. Staff seeking approval encounter the policy’s requirements as part of the request, rather than being expected to locate and interpret separate guidance independently.
Other participating councils have introduced different arrangements. One authority uses an AI Ethics Board to review proposed applications across its services, while another has implemented technical controls restricting access to unapproved AI systems. The government research does not identify those two councils.
These mechanisms address a weakness in relying exclusively on published policies. Employees may understand that their organisation has rules governing AI without knowing which services are approved, what information can be entered into them, or who must authorise a new application.
Connecting policy to software access and procurement can help councils assess proposed uses before they become embedded within routine work. Such checks may examine information security, data protection, supplier arrangements, and the operational consequences of using automated outputs.
Technical restrictions offer a different form of control by limiting access to particular services. They may prevent some unauthorised use, but cannot by themselves establish whether an approved application is being used appropriately or producing dependable results.
Ongoing oversight remains necessary where AI systems are connected to operational information, particularly when their outputs influence work affecting residents.
Oversight extends across council departments
The research found that AI policy development commonly involved specialists from information governance, IT, digital services, analytics, human resources, finance, legal teams, content design, and trade unions.
Devon County Council has established an AI Steering Group and AI Board with representatives from different parts of the organisation. Their remit includes oversight of AI activity and consideration of the ethical issues associated with proposed applications.
Involving several departments can help identify requirements that might otherwise be missed. A technology team may understand the security implications of connecting a new system, while service managers can explain how the application would affect established decisions and interactions with residents.
Information governance specialists may need to consider which records the technology can process, and legal advisers may examine obligations arising from its intended use. Employee representatives can contribute to discussions about changes in working practices and the responsibilities of staff expected to supervise automated processes.
The government’s interviews do not establish that any one governance structure is more effective than the alternatives. They describe arrangements adopted by particular authorities and identify common considerations for councils developing their own systems.
Existing software procurement and information management processes can provide a foundation for AI controls, although additional review may be needed where an application generates outputs that are not wholly predictable or operates across several services.
Training connects policy with daily decisions
Councils participating in the research reported using staff bulletins, management briefings, workshops, videos, internal guidance pages, learning pathways, and designated AI champions to explain their policies.
Leeds City Council has incorporated AI into mandatory information governance training and developed an internal 45-minute course. It has also created dedicated intranet guidance and additional learning resources for employees.
These activities respond to the difficulty of turning a formal policy into decisions made by individual staff. Employees must recognise when a proposed use requires approval, understand the limits of permitted tools, and know how to raise concerns about an output or an unfamiliar application.
Training alone does not establish that AI systems are secure or reliable, however. Organisations still need appropriate testing, access controls, supplier assessment, and procedures for reviewing consequential decisions.
The research also highlights transparency towards residents. Councils are considering when they should explain that AI has contributed to a service, communication, or decision, and what human oversight and alternative routes should be available.
The appropriate explanation may differ between an internal administrative task and an application whose output substantially affects a resident. Developing those arrangements before a service is introduced can make it easier to preserve accountability and provide meaningful information to people using it.
Common resources remain in development
The Local AI team is using its interviews and a review of more than 40 existing council policies to develop a shared template. The intention is to provide practical material covering responsibilities and governance requirements while allowing authorities to adapt it to their own structures and circumstances.
Alongside the template, the government is co-designing a Responsible AI Charter that would establish a common baseline of commitments. Workshops with councils have already contributed to development of both resources.
Neither publication is currently a mandatory national policy for local government. Councils will continue to require arrangements appropriate to their services, existing governance systems, and the risks associated with individual applications.
The work takes place as AI becomes more prominent in local government operations. Recent deployments of planning-related AI across 15 councils illustrate a separate development in the application of the technology to public services.
As more applications move into operational use, councils will need to consider not only the technology’s capabilities but also who authorises its use, what records it can access, how outputs are checked, and how errors are corrected.
The Local AI team is continuing to work with authorities on the policy template and charter, with publication expected by spring 2027. The November research provides examples of emerging governance practice, but does not yet measure how effectively those arrangements reduce incidents or improve public service outcomes.












