AI, Transparency and the New Rules of Public Decision-Making

Last month the RED newsletter noted the imminent adoption of information standards

and the increasing prevalence of AI in planning. This month we take a look at how

information regulations brought in more than 20 years ago could be used to force

disclosure around the use of AI – and other governance issues bubbling under the

surface.

Your AI Trail Is Showing: Environmental Information in the Digital Age

If you thought the Environmental Information Regulations 2004 (EIR) were a relic of a

more paper-heavy era, think again. Far from gathering dust, they are increasingly being

used as a powerful lever to expose digital records—emails, spreadsheets, draft reports

and even interactions with AI tools.

For anyone working in planning, policy or environmental decision-making, the message

is clear: if it touches the environment, someone, somewhere, can probably ask to see

it. And increasingly, they may be compelled by the courts to do so.

A regime built on disclosure

The origins of EIR lie in the 1998 Aarhus Convention and the EU Directive 2003/4/EC on

Public Access to Environmental implemented in England through the Environment

Information Regulations SI 2004/3391 (“EIR”). But what matters in practice is how

expansive the regime is.

EIR starts from a simple premise: disclosure is the default. In other words, you don’t

start from “why should we disclose this?” but from “can we really justify not doing

so?” Regulation 12(2) creates a presumption in favour of disclosure, and the courts

have repeatedly emphasised that exceptions must be interpreted narrowly. As the

courts made clear in Vesco v Information Commissioner, there are “more hurdles to

jump before a public authority may legitimately refuse a request for environmental

information.”

“Environmental information” is defined with deliberate breadth, covering not just the

obvious—air, water, soil and biodiversity—but also the things that influence them:

policies, plans, emissions data and impacts on human health and built structures.

Crucially, it applies to “any information in written, visual, aural, electronic or any other

material form”. In a 2021 case (Committee on Climate Change v ICO) a member of the

public asked for the spreadsheets relied on by the Committee to produce the Net Zero

Report. The Committee refused the request but the court required disclosure, ruling

that 1) the spreadsheets were not “unfinished documents”; 2) the spreadsheets should

be disclosed even if their calculations were not used in the report itself and 3) there

was “an extremely strong public interest in enabling scrutiny of the data, models and

calculations” relied on.

Nowhere to hide

One of the most relied-upon exceptions under EIR is the “internal communications”

exemption—the space where options can be tested and advice freely given. But even

that space is shrinking under scrutiny. In the 2025 case of Amin v The Information

Commissioner Mr Amin asked for disclosure of the internal government

correspondence relating to the grant of planning permission for a coal mine near Whitehaven in Cumbria. The High Court had already ruled that the

permission should be quashed and had also refused to order disclosure of those

documents, but the tribunal took a different approach, considering that there was a

very strong public interest in disclosure and ordering the documents to be disclosed.

The takeaway is uncomfortable but clear: calling something “internal” does not make it

immune to EIR.

AI and The new normal?

The EIR regime was designed in a pre-digital era, but it has proved surprisingly

adaptable. The disclosure by Department for Science Innovation and Technology of

the then Minister’s ChatGPT history in response to an FOI request suggests that other

public officials could be required to make similar disclosures, especially under the

more onerous standards of EIR. And in Elsbury v The Information Commissioner [2025]

UKFTT 915 (GRC) the tribunal required HMRC to disclose details of the Large Language

Models it was using, including their purposes, how they were selected, the impact of

their use and plans for future use. The Tribunal stated that: “We also agree with the

Appellant's argument that transparency on HMRC's part is particularly important when

AI's role in decision-making is a pressing concern globally.” This wasn’t a planning

case, but its implications are hard to ignore: if AI is being used to draft, analyse or

advise on a planning issue, it will come within the scope of EIR.

Over privileged?

Regulation 12(5)(b) allows information to be withheld where disclosure would

“adversely affect” the course of justice, the ability of a person to receive a fair trial or

the ability of a public authority to conduct an inquiry of a criminal or disciplinary

nature, which extends to legally privileged information. But as the Chancellor of the

High Court has recently noted, once AI is used by either party “one cannot assume that

confidentiality is preserved” and this regulation might be irrelevant. This has been

confirmed by the Upper Tribunal (albeit in the immigration context) in R (Munir) v

SSHD [2026] UKUT 81 (IAC), which concluded that: “Uploading confidential documents

into an open-source AI tool, such as ChatGPT, is to place this information on the

internet in the public domain, and thus to breach client confidentiality and waive legal

privilege”. And the SRA has gone one stage further, issuing a formal warning notice

noting that once client materials were entered into an open source AI tool, then “legal

professional privilege may be permanently waived and unable to be recovered.”

The Hidden Risks of AI in Public Decisions

There are of course other risks to be mindful of when using AI tools, particularly in the

sphere of decision-making, and especially when those decisions are public. Developers

and decision-makers alike will need to consider a range of matters such as automation

bias, bad data, hallucination and data security. The use of AI tools will not dispense

with basic principles of lawfulness and fairness, including (for example) not fettering

one’s discretion, the duty to give reasons or the public sector equality duty (“PSED”).

Examples of things going wrong from other fields serve as a cautionary tale. In R

(Bridges) v South Wales Police [2020] EWCA Civ 1058, the Court of Appeal found that

the Police had breached the PSED because they did not take reasonable steps to

enquire whether the automated facial recognition technology it was using was free

from bias in respect of race and gender (even though there was no finding of bias in

relation to the software). In R (Ames) v Lord Chancellor [2018] EWHC 2250 (Admin),

the Legal Aid Agency was found to be under an obligation to disclose its fee

“calculator” for counsel fees, without which there could be no transparency about the

figures being generated, and which amounted to procedural unfairness. During COVID,

Ofqual had to scrap its system relying on a standardisation algorithm to award A-level

and GCSE grades for students who had been unable to sit exams that year, due to

threatened legal challenge.

At least one way of mitigating such legal risks is to ensure clear policies are in place

which govern the use of AI and digital tools in decision-making and the discharge of a

public body’s functions. It’s not unusual for WhatsApp to be used to facilitate team

communications, but recently the Metropolitan Police came under scrutiny for doing

exactly that, due largely to the function that deletes conversations after 90 days. In

this case the use was governed by specific Met guidance, and while the legal challenge

to the policy itself was dismissed, the Court had reservations about the guidance, such

as the deletion of potentially relevant material in the context of criminal proceedings.

None of these issues should stop any profession or organisation from using AI – in fact

we are indebted to Copilot for turning some dry legal notes into the first draft of this

article and adding some attention-grabbing headings. But do think about the

consequences before you press send, save that spreadsheet or ask Copilot for a

helping hand. AI may be moving fast, but the principles of transparency, accountability

and good decision-making are not being left behind.