Can councils use AI to process planning applications?

Yes, for the paperwork around the decision: summarising objection sets, first-drafting officer reports, checking applications for completeness and searching policy. No, for the decision itself. A planning application must be determined by the authorised officer or the planning committee weighing the material considerations, and an AI system cannot lawfully take that role. The workable pattern is AI that drafts and summarises inside the council's own boundary, with a record showing exactly what the officers read and what they decided.

The question matters in 2026 because planning departments sit under statutory determination periods, in England eight weeks for most applications and thirteen for major development, while carrying chronic backlogs and a workload that is almost entirely documents: application forms, design and access statements, consultee responses, and public objections that arrive in the hundreds for contentious schemes. The pressure to adopt AI is real. So is the risk of adopting it in a way that a judicial review later takes apart.

What can AI genuinely do in a planning department?

Four things, all of them assistance rather than judgement.

• Summarising objection sets: grouping hundreds of representations into themes, with counts of who raised what.
• First-drafting officer reports: assembling site history, policy context and consultation responses into a working draft for the case officer.
• Completeness checking: testing a submission against local validation requirements before it enters the queue.
• Policy and precedent search: finding the relevant local plan policies and comparable decisions across the authority's own record.

Each of these compresses hours of reading. Each also carries the same failure mode: a summary can bury a material planning consideration inside an aggregate. The discipline that answers it is simple: summaries are navigation aids, and the case officer reads the underlying representations wherever a theme or an outlier signals something material.

Can an AI lawfully make the planning decision?

No. Determination rests with the decision-maker the law recognises: the authorised officer acting under the scheme of delegation, or the planning committee. The decision-maker must weigh the material considerations and reach a planning judgement, and that judgement cannot be handed to a system. The reworked UK automated-decision regime under the Data (Use and Access) Act 2025 adds a data protection dimension, but planning law gets there first: a decision that was in substance produced by a machine, with an officer signing what the machine concluded, is a decision waiting to be challenged. Residents notice when reasons read like templates, and their advisers know what to do with that.

What about the personal data inside objections?

Objections are not neutral documents. They contain names and addresses, and in contentious cases they contain allegations about neighbours: noise, behaviour, business activity, sometimes safeguarding concerns. UK GDPR applies to all of it. Sending an objection set to a public cloud AI service is a disclosure of that personal data to a third party the objector never anticipated, on retention terms the council does not control. The same summarisation running inside infrastructure the council owns raises none of those questions, because the data never leaves the authority's boundary.

Will an AI-assisted decision survive judicial review?

It can, if the record is right. Judicial review probes how the decision was reached, and the reasoning trail must show human judgement on the material points. A useful standard is the decision file test: could the council show a court what the AI summarised, what the case officer read in full, and where the officer's own reasoning engaged each material consideration. A council that can answer those three questions from a verifiable record is in a strong position. A council reconstructing the answer from mutable application logs and memory is not. Freedom of information and environmental information rules reach these records too, so the working should be written on the assumption that it will one day be read.

What does a sovereign deployment look like for a council?

AI running on hardware the council owns, inside its own network, with no outbound route for the data. On Mickai, a Sovereign Intelligence Operating System, the work runs behind a zero-egress perimeter, every action is sealed to a post-quantum signed audit ledger bound to hardware-attested identity, and the record verifies offline, so a court, an auditor or an ombudsman can check what happened without taking anyone's word for it. Inference is selectable between CPU and GPU, which matters for authorities whose estate was not bought with accelerators in mind.

How should a planning team start?

With decided cases, not live ones. Take a set of determined applications, run the summarisation and drafting against them, and compare the output with what officers actually produced, against criteria agreed before the exercise starts. Move to live casework in stages, the lowest-risk step first, completeness checking, then summarisation with officer review, and keep report drafting under full case officer ownership throughout. The pattern is boring by design: planning decisions get challenged, and boring survives challenge.

“AI can compress the paperwork around a planning decision, but the judgement inside the decision must remain visibly human.”

How the boundary, the sealed ledger and the wider architecture fit together is set out at /sovereign-ai, and the film at /film shows the interface working through document sets of exactly this kind.

Frequently asked questions

Can we use ChatGPT to summarise public objections to a planning application?

Objection sets contain personal data and often allegations about identifiable neighbours, so entering them into a public cloud service is a disclosure to a third party the objectors never anticipated, and one the council would have to defend under UK GDPR. Summarisation is valuable, and it should run inside infrastructure the council controls, where no such disclosure occurs.

Does using AI make a planning decision unlawful?

Not in itself. The law requires determination by the authorised officer or committee weighing the material considerations, and AI that drafts or summarises does not change who decides. The risk arises when the decision is in substance the machine's output, signed rather than made. The record should show the officer's own engagement with the material points.

What happens if an AI summary misses a material consideration?

The failure lands on the decision, which is why summaries must be treated as navigation rather than substitutes. Process should require the case officer to read the underlying representations wherever a theme, a count or an outlier suggests something material, and a sealed record showing both the summary and what the officer read makes that discipline provable.

What is the decision file test for AI-assisted planning decisions?

Three questions, answered from the record: what did the AI summarise, what did the case officer read in full, and where did the officer's own reasoning engage each material consideration. A council that can answer all three from a sealed, verifiable record can defend the decision. A council that cannot is relying on memory.

How should a council pilot this without risking live applications?

Run the AI against already-determined applications and score its summaries and drafts against what officers actually produced, with pass criteria agreed in advance. Promote it to live use in stages, completeness checks first, and keep every live output under officer review until the comparison record justifies going further.