Thred

Accounting firm AI policy: a template for UK practices, clause by clause

Manoj Tutika · Published 16 November 2026 · Last checked 16 November 2026

Working on this yourself? Talk to Mark, our co-founder. We’re here to solve it with you.

Book a demo

Why does an accounting firm need an AI policy?

Because your staff already use AI. In Karbon's State of AI in Accounting Report 2026 (593 respondents, surveyed October 2025, mostly in the US), 98% of respondents used AI, and only 21% of firms had an AI policy. The gap between the two is where client data ends up in the wrong tool, unchecked output ends up in a client letter, and nobody can say afterwards which was which.

The rules themselves are not new. ICAEW applies the five fundamental principles of its Code of Ethics to generative AI: be open about AI use, keep confidential client information out of public tools, apply scepticism to the output and keep accountability with the accountant. Its updated guidance on professional conduct in relation to taxation says the same for tax work: AI does not replace professional judgement. UK GDPR still applies to every piece of client personal data. A policy turns those principles into instructions a junior can follow on a Tuesday afternoon.

The template

Copy the clauses below into your own document and replace the words in square brackets. Keep it short enough that people read it.

  1. Scope. This policy applies to every partner, employee, contractor and subcontractor of [Firm name] who uses any AI tool for firm or client work, on firm or personal devices. AI tool means any software that generates text, numbers, code or decisions from a model, including chat assistants, AI features inside our practice and accounting software, and AI agents that read or write client ledgers.
  2. Approved tools. Only tools on the approved list in Annex A may be used for client work. Each entry states the tool, the plan or edition, what it may be used for and the highest data class allowed. New tools are added by [AI owner] after a check of the supplier's terms, data location, training-on-data settings and, where client personal data is involved, a data processing agreement and a DPIA. Personal and free accounts are not approved for client work.
  3. Client data. Staff classify data before using it with AI. Public data may go into any approved tool. Firm internal data may go into tools approved for it. Client confidential data, including personal data, bank details, payroll and tax records, may only go into tools approved for that class, and only the minimum needed for the task. Special category data and data of clients who have opted out of AI processing never goes into an AI tool.
  4. Confidentiality. Our duty of confidentiality applies unchanged. Staff do not paste confidential client information into public generative AI tools, even if parts of it are public elsewhere. Outputs containing client information are stored only in the client file.
  5. Review and sign-off. AI output is a draft. A qualified person who understands the work reviews it before it is relied on, sent to a client, filed or posted to a ledger. The reviewer checks figures against source records, checks that current tax rules, thresholds and reliefs were applied, and signs off in the usual way. Responsibility for the work stays with the engagement partner. AI agents that write to a client ledger do so only through approval steps or rules set by [role], recorded per client.
  6. Disclosure to clients. Our engagement letters state that we use AI tools in delivering our services, which categories of tools and how client data is protected. Where AI has materially shaped advice, we say so. We honour a client's written request not to have their data processed by AI.
  7. Logging. We keep a record of which approved tools are used on which clients and for which tasks. Agents that act in client ledgers must keep an audit trail of every read and write, the source document and who approved it. Logs are retained for [period] in line with our record-keeping policy.
  8. Training. Everyone completes AI training before using an approved tool for client work, and a refresher every [12] months. Training covers this policy, data classification, prompting for accounting work, typical failure modes such as invented figures and outdated rules, and how to review output.
  9. Breaches. Anyone who puts client data into an unapproved tool, relies on unchecked output or sees an AI error reach a client reports it to [AI owner] the same day. We assess whether it is a personal data breach, which under UK GDPR Article 33 must be reported to the ICO without undue delay and, where feasible, within 72 hours, unless it is unlikely to result in a risk to individuals, whether the client must be told, and whether our professional indemnity insurer needs notice. Breaches of this policy are handled under our disciplinary procedure.

Annex A is a simple table: tool, plan, approved uses, highest data class, data location, DPA in place, DPIA date, owner.

Which decisions does the firm have to make first?

A template leaves the hard choices to you. Make these before you publish the policy, write the answers into it and review them every year.

The decisions behind an accounting firm AI policy
DecisionOptionsWhat to weigh
Who owns the policyManaging partner, compliance partner, operations leadNeeds authority to say no to a partner's favourite tool
Which tools are approvedAssistants on business plans, AI inside practice software, ledger agentsSupplier terms, training-on-data, data location, DPA
Data classes and limitsTwo, three or four classesSimple enough to apply without asking
Personal accountsBanned for client work, or allowed for public data onlyMost leaks start here
Review standardEvery output reviewed, or sampling for low-risk tasksRisk of the task, seniority of the user
Agent write accessRead-only, approve each write, rules per taskStart read-only, widen per client as accuracy is shown
Client disclosureEngagement letter clause, plus case-by-case disclosureICAEW expects openness about AI in client work
Client opt-outAllowed, or not offeredCost of running two processes
DPIAPer tool, per use caseUK GDPR Article 35
Log retentionMatch working papers, or shorterRegulator reviews and disputes
TrainingOne-off, annual, per new toolRole-based training saves time

How do you roll the policy out without stalling work?

  1. Inventory what people use today. Ask, do not assume. You will find personal accounts.
  2. Approve two or three tools that cover most needs, so there is a safe option on day one.
  3. Publish the policy with Annex A, run a one-hour session, and collect a signed acknowledgement.
  4. Update the engagement letter wording at the next renewal round.
  5. Review after three months with real incidents and questions, then once a year.

For a structured first trial, see the four-week AI pilot for accounting firms, and for the risks of connecting assistants to client books, see connecting Claude or ChatGPT to your accounting system.

How do Thred's agents fit a firm's AI policy?

Thred is a team of AI agents that handle daily work for finance teams and accounting firms, on your existing accounting software, under the rules your team sets. The policy clauses above map directly onto how Thred works in a practice.

In Thred's firm portal, one chat works across all your client administrations. The Bookkeeper keeps each client's record up to date. The Accountant handles filings and optimises for taxes. The Connector finds context in your connected tools. The Collector, The Payer, The Treasurer, The Analyst and The CFO cover receivables, payables, cash, variance and finance leadership for clients who need them. On Exact Online the Bookkeeping Quality and Close agents run per client on a schedule.

Each client can be set to read-only or to ask before writing, which is clause 5 in practice. Every write is an approval card showing the client, accounts and amounts until you set a rule, and every action is recorded, which is clause 7. Data is hosted in AWS Frankfurt with a data processing agreement in every contract, which is the evidence clause 2 asks for. Agents also work on Xero and Moneybird. See Thred for accounting firms.

Frequently asked questions

Is there an AI policy template for accountants?

Yes, this page has one. It has nine clauses: scope, approved tools, client data, confidentiality, review and sign-off, disclosure to clients, logging, training and breaches. Copy it into your own document, fill in the choices in the decisions table and have a partner own it.

What should an AI acceptable use policy for an accounting firm cover?

At minimum it should cover which tools are approved, what client data may go into which tool, who must review AI output before it is relied on, how AI use is recorded, what staff must be trained on, and how a mistake or data incident is reported. Everything else is detail.

What does ICAEW guidance say about AI?

ICAEW applies the five fundamental principles of the Code of Ethics to AI. In short, be open about AI use in client work, do not put confidential client information into public tools, apply professional scepticism to output, and keep accountability with the accountant. Its updated tax conduct guidance (PCRT) takes the same line.

Can I put client data into ChatGPT or Claude?

Not into a public consumer account. ICAEW's guidance says confidential client information should not go into public generative AI tools. A business or enterprise plan under a data processing agreement, with training on your data switched off, is a different case, and your policy should name exactly which plans are approved.

Do we need a DPIA before using AI on client data?

UK GDPR Article 35 requires a data protection impact assessment before processing that is likely to result in a high risk, in particular when using new technologies. Many firms treat AI tools that touch client personal data as needing one. Do it once per tool and use case, and keep it with the policy.

How many accounting firms have an AI policy?

In Karbon's State of AI in Accounting Report 2026, based on a survey of 593 people in October 2025, mostly in the US, 21% of firms had an AI policy, although 98% of respondents used AI. Most practices are using AI ahead of writing down how.

Sources

  1. ICAEW, Generative AI guide, generative AI and ethics
  2. ICAEW, AI and accountants, the rules and guidance you need to follow (20 May 2026)
  3. UK GDPR Article 35, data protection impact assessment (legislation.gov.uk)
  4. UK GDPR Article 28, processors (legislation.gov.uk)
  5. UK GDPR Article 33, notifying a personal data breach (legislation.gov.uk)
  6. Karbon, The State of AI in Accounting Report 2026

Read next

Where Thred comes in

Accounting Firms

Make your firm more efficient with AI, so you have more time for relationships, new customers and advisory work.

Finance, handled.

Book a demo, or request a callback and we will call you.

Book a demo