Technology & Innovation

AI ethics and compliance: A practical guide for accountancy practices

The PCRT bodies published AI guidance in January 2026, and CCAB has just opened an ethics hub. Here is what both mean for your practice in daily terms.

Published 17 August, 202610 min read Share

Key takeaways

• The seven PCRT bodies published topical guidance on the ethical use of AI tools on 19 January 2026, alongside an updated PCRT effective from 1 January 2026.
• The guidance introduces no new rules. It reads the five existing fundamental principles through an AI lens.
• You remain responsible for any work your practice produces, whether or not an AI tool helped create it.
• CCAB has launched an AI hub with a draft Statement to the Profession on the Ethical Use of Artificial Intelligence, due for discussion at an online event on 25 September 2026.
• PCRT is reflected in HMRC’s Standards for Agents, which makes it the practical benchmark for tax work across the profession.

Most accountants asking whether they are allowed to use AI have already had their answer.

The live question now is how to use it in a way that would survive a complaint, a client challenge, or a disciplinary review.

Part, if not all, of that answer arrived in January 2026, when the seven professional bodies behind Professional Conduct in Relation to Taxation (PCRT) published topical guidance on the ethical use of AI tools.

It came alongside an updated PCRT effective from 1 January 2026, and it has since been joined by a new CCAB hub dedicated to ethical AI use.

None of it introduces new rules, which is the reassuring part and also the part practices tend to misread. The existing principles apply in full, and applying them to AI takes deliberate work.

This short article is one attempt to suggest what that work looks like day to day, as follows:

What changed in January 2026, and what did not

Two things happened close together.

An updated PCRT took effect on 1 January 2026, and on 19 January the seven bodies that jointly produce it published topical guidance covering the application of PCRT to the ethical use of artificial intelligence tools.

The seven bodies are AAT, ACCA, ATT, CIOT, ICAEW, ICAS, and STEP. PCRT is reflected in HMRC’s Standards for Agents, which is why it functions as the benchmark for tax work across the profession rather than as one body’s house rules.

The guidance is deliberately not technical.

It does not tell you which tools to buy or how to configure them.

It interprets the five existing fundamental principles through an AI lens, identifies the ethical risks that arise under each, and offers safeguards against them.

The line that matters most is the one about responsibility. Members remain responsible for any work they produce, and regulated firms for any work the firm prepares, irrespective of whether AI tools were used in creating it. There is no partial defence available for an error the software made.

The five fundamental PCRT principles through an AI lens

Here they are in summary—but you owe it to yourself and your practice to review these fully (and get advice from your membership organisation or an expert source if in any doubt):

• Integrity: Be transparent about how AI supports your work. If a tool assisted with analysis or drafting, you should still understand the output well enough to explain it clearly to a client who questions it.
• Objectivity: The guidance cautions specifically against over-reliance. An AI suggestion should be challenged and considered against the client’s actual circumstances rather than accepted because it reads convincingly. Fluent output is not evidence of correct output.
• Professional competence and due care: You need enough understanding of a tool to supervise it properly. That includes knowing what it is poor at, which is usually more useful than knowing what it is good at.
• Confidentiality: Your client data obligations carry across unchanged. What enters a tool, where it is stored, and whether it trains a model are now professional questions rather than purely technical ones.
• Professional behaviour: An AI-caused error damages your reputation in exactly the way any other error would, and rather faster, because clients are primed to notice this particular kind of mistake.

Which tasks should stay with a human

The useful test is not whether AI can do the task.

It is whether being wrong would be expensive and hard to detect.

Keep these with a qualified person:

• The final tax position, and any borderline or contentious treatment
• Any judgement requiring assessment of a client’s intentions or commercial rationale
• Going concern, materiality, and other matters of professional judgement
• Advice on arrangements where the boundary between planning and avoidance is in play
• Anything issued to a third party as the firm’s position
• Every sign-off, without exception

AI is genuinely useful for first drafts, summarising long documents, extracting data, flagging reconciliation differences, producing meeting notes, and shortlisting research avenues.

One area deserves particular care.

Research from AccountingWEB in association with Sage reveals around half of accountants who use AI tools outside their core software use them to research tax rules or other technical knowledge, making it the single most common external use.

It is also the riskiest. A confident and wrong answer on a tax point is difficult to spot without checking the source anyway.

Treat an AI answer on a technical matter as a lead to verify against legislation, HMRC manuals, or your usual research tool. Never treat it as an authority in itself.

Telling clients when AI has been involved

The integrity principle requires transparency about how AI supports your work.

The practical question is, of course, how much to tell them.

Under-disclosure creates the risk that a client discovers AI involvement in a way that feels concealed. Over-disclosure, particularly a lengthy technical explanation attached to routine work, tends to alarm people who were not worried in the first place.

A proportionate approach has three parts.

1. Add a short clause to your engagement letters describing how AI may be used and confirming that qualified people review all work.
2. Publish a page on your website carrying the same information in plainer language.
3. Where AI has materially shaped a specific deliverable, say so in the document itself.

Ask the question in the other direction too. Some larger clients now include AI clauses in their own supplier terms, and a few restrict the use of external models on their data entirely. Far better to find that out at engagement than at review.

The black box problem, explainability, and audit trails

The AccountingWEB research reveals that around one in eleven accountants name a loss of transparency and visibility as their single biggest AI concern.

It is a minority worry, and it sits underneath the much larger accuracy concern, because an output you cannot explain is an output you cannot properly check.

The standard to apply is the one you already apply to a spreadsheet with a hidden formula—if you cannot show how a figure was arrived at, you cannot defend it to a client, a regulator, or a tribunal.

What to record, as a minimum:

• Which tool and which version produced the output
• What was submitted to it
• What it returned
• Who reviewed it, what they changed, and when
• The final human sign-off

Keep the review step inside the workflow rather than in somebody’s memory. A sign-off field in your practice management system does the job. Align retention with your normal file retention policy so there is nothing separate to remember.

Audit firms have a further layer to consider, as the Financial Reporting Council has issued guidance addressing the risks and benefits of generative and agentic AI in audit work.

Writing an ethics statement clients will actually read

Your AI ethics statement is a client-facing document, so write it like one.

Cover five things: What you use AI for, what you do not use it for, what happens to client data, who checks the work, and finally, who to contact with a question.

Keep it to a page and write it in plain English. Hedged, heavily qualified language reads as evasion to a client even when it is legally careful.

Review it annually and whenever you approve a new tool.

This is a moving area and there is more coming. CCAB, which brings together ICAEW, ACCA, ICAS, CIPFA, and Chartered Accountants Ireland, has launched an AI hub gathering guidance, checklists, podcasts, and resources on ethical AI use.

Its ethics group has produced a draft Statement to the Profession on the Ethical Use of Artificial Intelligence along with six case studies, which will be discussed at an online event on 25 September 2026. That is worth putting in the diary before you finalise your own wording.

Final thoughts

The reassuring conclusion from the PCRT January 2026 guidance is that nothing about your professional obligations has changed. The demanding conclusion is the same sentence, read the other way round.

Nonetheless, when it comes to guidance, this is still an evolving world, and it will pay over the coming months and years to pay attention to your membership body’s documentation and training in this area.

AI can support the work, and responsibility for that work stays exactly where it always sat. Every principle you already apply carries over, and applying them well now requires you to understand tools that are unfamiliar and moving quickly.

Treating AI as a firm-wide risk management matter, aligned across engagement terms, internal controls, and daily behaviour, works considerably better than treating it as an IT decision.

Your takeaway: Read the PCRT topical guidance, and use the accompanying webinar to cover the change in your 2026 CPD. Write down which tasks in your practice must stay human. Add an AI clause to your engagement letters. Start logging tool, input, output, and reviewer on anything client-facing. Put 25 September in the diary and see what CCAB publishes before you finalise your own statement.

None of this is onerous. All of it is the difference between using AI confidently and using it with hope.

Read below—State of the nation: AI in accountancy and bookkeeping, produced by AccountingWEB in association with Sage

Frequently asked questions

Are accountants allowed to use AI under PCRT?

Yes. The seven PCRT bodies published topical guidance on 19 January 2026 confirming that AI tools may be used in tax work, provided the existing fundamental principles are applied. The guidance introduces no new rules and instead interprets integrity, objectivity, professional competence and due care, confidentiality, and professional behaviour through an AI lens. Members remain fully responsible for any work produced, whether or not AI was involved in creating it.

What did the January 2026 PCRT AI guidance actually say?

It applies the five existing PCRT fundamental principles to the use of AI tools, identifies the ethical risks arising under each, and sets out safeguards to address them. Key points include being transparent with clients about AI involvement, challenging AI outputs rather than accepting them at face value, maintaining enough understanding of a tool to supervise it, and protecting client confidentiality. It is ethical guidance rather than technical implementation advice.

Do accountants have to tell clients they use AI?

The integrity principle requires transparency about how AI supports your work, so some form of disclosure is expected. A proportionate approach is a clause in your engagement letter describing how AI may be used and confirming human review, a plain-English page on your website, and a specific note where AI has materially shaped an individual deliverable. It is also worth asking clients whether their own supplier terms place restrictions on AI use.

Who is responsible if an AI tool makes a mistake in tax work?

You are. The PCRT guidance is explicit that members remain responsible for any work they produce, and regulated firms for any work the firm prepares, irrespective of whether AI tools were used in creating it. There is no partial defence for an error originating in software. This is why human review, documented sign-off, and verification of technical outputs against primary sources matter so much.

What should be in an accountancy practice’s AI ethics statement?

Five elements, on a single page and in plain English. What you use AI for, what you do not use it for, what happens to client data, who reviews the work, and who a client should contact with questions. Review it annually and whenever you approve a new tool. CCAB’s AI hub offers guidance, checklists, and case studies you can adapt, with a draft profession-wide statement due for discussion in September 2026.

Browse more topics from this article

• Accountant accounting
• AI
• Digital transformation
• How to grow your accounting practice
• Workplace technology