LEGAL MATTERS - AI in Local Government
By Francesca Pridding, partner, LGRC
I have twin daughters, one of whom is fundamentally opposed to the use of AI on environmental and ethical grounds. The other is an estate agent, with arguably fewer scruples, and happily embraces the time she gets back from no longer having to write eye-catching adverts and re-word clichéd property descriptions. Debates rage in my house on a regular basis.
I had been contemplating that the average clerk might sit somewhere between these two positions. Are you looking on in mild panic, anticipating the havoc the bots will wreak when they get their hands on your allotment plan? Or are you diving head-first into a sea of vibe-coding, yelling ‘hallelujah, I can run this council from a beach in the Bahamas, so long and see y’all at conference!’?
Today’s reality, thanks to some dire warnings from owners of the largest AI companies, means we’re now at the stage of ‘heck, it’ll all be over for the human race by 2030, so we’d best be polite to our electronic overlords!’
In the meantime, we probably need to understand that AI in some form is going to be here to stay, so we need to understand where we can get AI working smart for our councils and where we need to exercise caution.
The clerk/CEO of a council is under greater pressure than ever to perform miracles. The role is diverse, expanding and pressured and the tools that help are increasingly ChatGPT, Copilot, Gemini, GovAssist and now NALC is getting in on the action with NORA (see page 20).
Clerking, however, is a mixture of distinct types of task, and there are some jobs that do lend themselves to the judicious use of AI and some that should only ever be undertaken with human judgment and care, due to the legal and ethical implications.
What’s easy to miss in all the noise about chatbots and prompts is that this was never really a technology question. It’s the same governance question you’ve been living with your whole career, just wearing a slightly flashier costume. If a piece of correspondence is confidential, it’s confidential whether you typed it yourself or asked a bot to have a go at the first draft. If a set of minutes needs to accurately reflect what was resolved, that requirement doesn’t relax because AI helped you get there faster. If a recruitment process needs to be fair, defensible, and free from discrimination, that’s just as true when an algorithm suggested the interview questions as when you wrote them yourself at your kitchen table at 11pm. The tool changes. The rules that already sat on your desk don’t.
That is good news, in a roundabout way. You don’t need to learn a whole new rulebook for the AI age. You already know what confidential means, you already know what a lawful recruitment process looks like, you already know your council’s formal record has to be accurate.
The job now is recognising when those same old rules are quietly in play, even though what’s doing the drafting has changed.
Getting a first draft of a resident letter, an agenda covering note, or a set of interview questions out of AI, then checking and editing it yourself, carries about the same legal exposure as asking a colleague to have a go first: minimal, provided you actually read what comes back.
The same goes for using AI to get your bearings on an unfamiliar bit of legislation or summarise a long consultation document before you go to the primary source. Treat it as background reading rather than the answer, and there’s nothing here a court or an auditor would ever look twice at.
Where the risk creeps up slightly is admin at scale, minute-taking support, template letters, FOI response drafting. It is still low exposure, if a human signs off before anything is sent or published. Remember though, the ‘AI wrote it, not me’ justification has never once worked as a defence, whether that’s a defamation claim over something an AI-drafted newsletter said about a local business, or disciplinary action if a letter that went out under your name turns out to contain something you should have caught.
If you’re drafting a statement or a social media post for the Mayor or chair to put out under their name, and the AI-assisted version turns out to be inaccurate or overtly partial, it’s the councillor who’s exposed to a standards complaint, not you, so that check-before-it-goes-out habit matters even more when you’re drafting on someone else’s behalf. You’re still the author in law, whatever actually typed the words.
Then there’s the middle ground. Newsletter copy, social posts, a livelier chair’s report, all fine, provided a human reads it before it goes out under the council’s name. The specific thing to watch here is defamation: AI can produce something that reads as a confident statement of fact about a person or a business when it’s actually invented, and if that ends up in something the council publishes, the council can be sued, exactly as if a clerk had written it deliberately.
The same complacency can have even greater consequences with two other tasks. First, summarising long documents. Skimming an AI summary to get your bearings on a 40-page consultation is fine. Basing an actual recommendation to members on that summary, without ever opening the original, risks the resulting decision being challenged by judicial review, on the grounds that members weren’t properly informed or didn’t have regard to something they should have considered.
Remember, the riskiest legal things we handle as clerks are planning decisions and employment matters. Developers will look at any way possible to compromise a decision that didn’t go their way and employees are entitled to rigorous and fair handling of their employment affairs.
Second, and this is the real one to watch: AI doesn’t know when it’s wrong. It will invent a case citation, misstate a deadline, or misquote legislation with exactly the same confidence it uses to get things right, and it won’t flag the difference for you.
Solicitors have already been hauled up by judges for submitting AI-generated citations to cases that simply don’t exist. A council report is no less exposed to the same trap, and if a decision is later challenged, the ‘AI told us that was the rule’ claim won’t hold up any better than any other unverified assumption. Check anything AI tells you as fact against a primary source before it goes near a decision, a resident, or a document with the council’s name on it.
Then there’s the list marked ‘avoid, full stop’. Putting personal or confidential information into a public AI tool, resident correspondence, HR case files, exempt planning detail, is the big one.
Free tools may train on what you type in, and you often can’t be sure where that data ends up or whether it meets UK GDPR requirements. Get this wrong and it’s a personal data breach, reportable to the Information Commissioner’s Office, who can fine the council and require public disclosure of what happened. Where the data involved belongs to a member of staff, mishandling it can also be raised as an aggravating factor if that employee later brings an employment tribunal claim. Evidence that the council didn’t handle their information properly rarely helps a council’s case.
Letting AI draft your statutory record unsupervised sits in the same bracket. Minutes, resolutions and financial regulations are legal documents with formal requirements, and an AI-smoothed version that quietly drops a required detail, or subtly misrepresents what was actually resolved, can leave the underlying decision open to judicial review or challenge on procedural grounds. This time it’s the council’s own formal record that’s the weak point, rather than a briefing document. You’re still the one legally responsible for its accuracy, whatever helped write the first draft.
Recruitment and HR carry a different, equally serious exposure. AI models are trained on vast quantities of existing text and data, which means they inherit the biases already baked into that material, and that material was, and largely still is, disproportionately male-authored and male-centred.
An AI trained on the world as it has been, rather than as it should be, will happily reproduce that imbalance at speed and at scale, in gender as much as in age, disability or background. Let it draft your shortlisting criteria or interview questions unchecked, and a rejected candidate with a protected characteristic has grounds for an employment tribunal claim for discrimination, where compensation is uncapped.
Signing the council up to a paid AI tool, or plugging one into your systems, isn’t a decision for an individual officer to make quietly on a Tuesday afternoon. It’s a procurement and information-governance decision like any other supplier choice, and needs to go through the council’s normal approval process under your Financial Regulations. Skip that, and you’re looking at an audit finding at best, and at worst a clerk or RFO personally exposed if something goes wrong with a tool nobody formally approved.
In practice, that means a short council AI policy, agreed by members, setting out what’s allowed and by whom; never putting personal or confidential information into a public tool without checking how it handles data first; treating everything AI produces as a draft to be verified, not a finished answer; and making sure a human is always the one who’s accountable for what finally goes out.
LGRC is a local government consultancy working across parish, community and town councils in locum clerk/RFO and governance consultancy roles, including governance reviews and HR investigations for a range of client councils.
Through LGRC, Francesca Pidding also offers mediation as a fast, confidential alternative to tribunal or court proceedings for councillor disputes, staffing grievances, and other conflicts that can otherwise drag on for months, an option still underused by parish and town councils but well worth considering before things escalate.
She holds a BA (Hons) in Law, an LLM, a PgDip in Legal Practice, CiLCA accreditation, and accredited mediation training, and built the Community Clerks’ Network, a Facebook group of more than 2,400 clerks. https://lgrc.uk/
Written by Francesca Pidding, partner, LGRC
As appeared in Clerks & Councils Direct, November 2026
© CommuniCorp