Aipedia: Bookkeepers

AI in bookkeeping: rules, risks and PI

Aipedia, by the Apex broking team · Last reviewed 2026-10-27

UK bookkeepers use AI for bank matching, receipt capture, coding, payroll and client communication, but no bookkeeping body has issued AI-specific rules. Existing duties on competence, confidentiality and records apply, and the bookkeeper remains accountable for AI output.

Part of: AI in Bookkeepers

In short

AI features may now be built into cloud bookkeeping software: smart bank matching, receipt capture, auto-categorisation, payroll automation, forecasting and drafting assistants. ICB and IAB have published no AI guidance, so their general conduct rules apply. AAT applies the PCRT AI guidance to members' tax work. Making Tax Digital for Income Tax, from April 2026 for sole traders and landlords over £50,000, puts AI-categorised records straight into quarterly HMRC updates. ICB requires at least £50,000 of PI, AAT sets fee-linked minimums and IAB requires cover for the services provided. None has addressed AI and PI, so policy wording and declared services decide.

How bookkeepers use AI

Many bookkeepers did not choose AI; it may have arrived inside the software they already used. Bank feeds now come with smart matching that learns from past choices. Receipt capture tools read supplier, date, VAT and total from a photo. Coding rules predict the nominal code for each transaction. Payroll systems run pay, RTI submissions and auto-enrolment with little manual input. Reporting tools write management commentary, and generative AI chatbots and assistants draft client emails.

An AAT Comment article (19 August 2026) described practitioners using smart-match bank rules, document capture, drafting assistants, chatbots and cash flow prediction day to day. The common thread is speed. A bookkeeper with many small clients can process far more work, and can add services such as forecasting, credit control or MTD submissions that would once have needed more staff.

The common risk is that review weakens as volume grows. When a reconciliation balances and a VAT return calculates, it is tempting to treat the work as done. AI errors are often systematic rather than one-off, so a single wrong rule can affect months of entries or several clients.

The rules: who says what about AI

Bookkeeping has no statutory regulator. Bookkeepers in business must be supervised for anti-money laundering by a professional body such as ICB or IAB, or by HMRC, and professional bodies set conduct and PI rules for their members.

ICB: no AI-specific guidance. Its Professional Conduct Regulations cover competence (s.12(a), s.36), absolute confidentiality (s.14), records (s.1, s.17, s.25(d), ss.34 to 35) and data protection (s.37, which still refers to the Data Protection Act 1998).
IAB: no AI guidance found. Its AML supervision requires proof of PI covering the services you provide.
AAT: applies the PCRT Topical Guidance on the Use of AI (January 2026) to tax work: an engagement letter statement on AI use (1.3), checking output like a junior colleague's work (3.4), ultimate accountability (3.5), no client data in public AI tools without consent (4.2), and an audit trail of tools used.
HMRC: mandatory tax adviser registration catches bookkeepers who deal with HMRC for clients; see GOV.UK.

The CCAB draft Statement on ethical AI (June 2026) applies to members of CCAB bodies, not to ICB, IAB or AAT members as such. For the detail, see the guides below and the hubs for accountants and tax advisers.

Making Tax Digital for Income Tax

MTD for Income Tax is a major change to bookkeeping work, and many practices may deliver it with AI-assisted software. According to GOV.UK guidance updated 26 March 2026:

Qualifying income	In tax year	MTD from
Over £50,000	2024-25	6 April 2026
Over £30,000	2025-26	6 April 2027
Over £20,000	2026-27	6 April 2028

Partnerships will follow, with no timetable set. Each quarterly update is only as good as the categorisation beneath it. See MTD for Income Tax and PI for the insurance angle.

The main risks

Systematic errors: learned rules repeat mistakes across periods and clients, which can turn one error into many claims.
Tax consequences: wrong data flows into VAT returns and MTD updates, leading to HMRC assessments, interest and penalties that clients may claim from the bookkeeper.
Scope creep: AI makes it easy to answer tax questions or offer forecasts beyond your competence (ICB s.12(a)) and beyond your declared services.
Confidentiality: client data pasted into public tools or held by capture services raises ICB s.14 and PCRT 4.2 issues.
Automated decisions: UK GDPR Articles 22A to 22D, inserted by section 80 of the Data (Use and Access) Act 2025 and in force from 5 February 2026, require safeguards where significant decisions about individuals are made solely by automated means, such as some payroll or credit decisions.
Vendor blame: software terms commonly limit vendor liability, so the client will look to you. AAT's president said in an AAT Comment interview that "Accountability does not transfer to the vendor when the processing does" (opinion, not policy).

No disciplinary decision against a bookkeeper involving AI was found. In a related tax case, Elden v HMRC [2026] UKFTT 41 (TC), an accountancy firm's AI-drafted submissions were described as "professional incompetence".

The PI position

Body	PI requirement
ICB	s.27(c): PI against professional negligence claims for public practitioners. Minimum £50,000, appropriate to the size of the business.
AAT	Greater of 2.5 times gross fee income or £50,000 (sole traders) or £100,000 (partnerships and companies); maximum required £1m above £400,000 fee income. Full civil liability, fully retroactive. Six years' run-off strongly recommended (AAT PII rules).
IAB	Proof of PI covering the services your practice provides. No minimum stated.
HMRC registration	No PI requirement.

None of these bodies has said anything about AI and PI. In practice, a client claim caused by an AI-assisted error is usually treated like any other negligence claim, subject to the policy terms, excess and limit. The gaps tend to be services added without telling the insurer, data incidents that belong under cyber cover, fines and penalties, and errors repeated across many clients that test aggregation and limits. See does PI cover AI mistakes and PI claims examples for bookkeepers.

Aipedia guides for bookkeepers

AI guidance for bookkeepers: AAT, ICB, IAB: what each body has and has not published on AI, and how ICB conduct rules apply.
AI bank reconciliation and smart matching: bulk-accepted matches, duplicated entries and misstated VAT.
AI receipt and invoice capture: misread amounts, dates and VAT, and ICB record-keeping duties.
AI auto-categorisation of transactions: learned coding rules and errors repeated across months or clients.
VAT returns from AI-coded ledgers: wrong VAT treatment, missed reverse charges and penalty claims.
MTD for Income Tax: AI and bookkeepers: quarterly updates built on AI-categorised records.
AI-assisted payroll for small businesses: RTI, auto-enrolment and automated decision safeguards.
AI credit control and invoice chasing: wrong customers chased, tone and data risks.
AI-written client emails for bookkeepers: advice beyond competence and pasted client data.
AI cash flow forecasts for small clients: reliance on predictions and what to promise.
AI ID checks and AML for bookkeepers: automated ID verification and the firm's own CDD duty.
AI-generated management reports: confident narratives built on miscoded data.
Answering client tax questions with AI chatbots: competence limits, hallucinations and declared services.
AI software errors: bookkeeper or vendor?: why accountability stays with the bookkeeper.
Does PI cover AI mistakes for bookkeepers?: ICB, AAT and IAB PI requirements and AI-related claims.

How Apex helps

Apex arranges PI for bookkeepers. A broker can review your cover against ICB, AAT and IAB requirements, your declared services and how your policy treats AI-assisted work.

Sources

Frequently asked

Do bookkeepers have to follow AI rules?

No bookkeeping body has issued AI-specific rules as at October 2026. ICB and IAB members follow general conduct rules on competence, confidentiality and records. AAT members doing tax work follow the PCRT AI guidance that AAT applies.

Is a bookkeeper liable for mistakes made by AI software?

Usually, yes, as far as the client is concerned. The client engaged the bookkeeper, vendor terms commonly limit liability, and professional rules require careful work. The question in a claim is whether a reasonable review would have caught the error.

Does bookkeepers' PI cover AI errors?

Usually, if the work is a declared service and no exclusion applies. ICB, AAT and IAB have not addressed AI in their PI rules, so the policy wording decides. Check services, cyber exclusions and aggregation.

When does MTD for Income Tax start?

From 6 April 2026 for sole traders and landlords with qualifying income over £50,000 in 2024-25, then from 6 April 2027 for those over £30,000 and from 6 April 2028 for those over £20,000.

Check your PI covers how you use AI

Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.

Start a PI quote Request a callback

Apex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.