§ InsightsNo. 113

Writing an AI use policy for an accountancy practice

A practical outline of an AI use policy for a UK practice, built on the ICAEW Code of Ethics, ICAEW guidance and the FRC's AI guidance.

An accountancy practice team discussing a draft AI use policy around a meeting table

An AI use policy for an accountancy practice is a short written statement of which AI tools staff may use, on what work, with what client information, and who checks the output before it reaches a client file. This article is for partners and managers who have to write one. It works through scope, approval of AI output, confidentiality under the ICAEW Code of Ethics, data handling, training, and review of the policy itself.

What the policy is for

An ICAEW Audit and Beyond article of 14 July 2025, written by a partner in a smaller firm about audit, advises firms to create a simple AI usage policy, and to keep it concise and focused on the key principles so that staff are clear. ICAEW’s topical guidance of 19 January 2026 on applying the Professional Conduct in Relation to Taxation (PCRT) standard to AI tools adds that a policy may outline the acceptable use of AI tools and staff disclosure where an AI tool has been used, which can help safeguard a firm against unknown or inappropriate use of AI.

That guidance is written for tax work. It rests on the fundamental principles in the ICAEW Code of Ethics, which apply to every professional activity, so we use it by analogy for accounts. Its reminder on supervision carries over directly: a member is responsible for the work done by staff and others under their supervision. A policy is how a practice shows it has set the rules for that supervision down.

A policy is also the starting point for our guide to signing off accounts that an AI tool has reviewed, which covers the responsibilities the policy has to respect. The table below lists the sections we would expect in a practice’s policy and the source behind each.

Policy sectionWhat it settlesAnchor
ScopeWhich tools and which work are coveredICAEW, concise key principles
Approved toolsWhat staff may use, and who approves a toolCode R320.11 and 320.11 A1
Approval of outputWho checks AI output, and how it is recordedICAEW PCRT guidance
Confidentiality and dataWhat may be entered into which toolCode subsection 114
TrainingWhat staff are taught, and how oftenCode R113.2
ReviewWhen the policy is revisitedOur recommendation

Scope and approved tools

Scope should cover any tool that uses AI, including features inside software the practice already has. The Audit and Beyond article describes starting with the AI capabilities embedded in existing tools, which is a reminder that staff can meet AI without installing anything new. A scope clause that lists only standalone products would miss them.

The Code gives a framework for approving a tool. Paragraph R320.11 requires an accountant who intends to use the output of technology to determine whether that use is appropriate for the intended purpose. The factors in 320.11 A1 include whether the accountant can understand, use and explain the technology, whether it has been appropriately tested and evaluated, the firm’s oversight of its design, implementation and updating, the controls over procedures for authorising user access and overseeing such use, and the appropriateness of the inputs, including data. Each factor can become a line in the approval record for a tool.

The FRC’s audit guidance offers a model for the register itself. It expects firms to document centrally the history of when versions of a tool were approved and retired, and it refers to a centrally maintained list of versions approved for use. That guidance is written for audit firms, but a practice’s policy can use the same idea: one list of approved tools with the date each was approved, the person who approved it, and the work it is approved for. Where a tool is bought in, our guide to questions to ask when choosing and testing an accounts review tool sets out what to ask the supplier before the policy approves it.

Approval of AI output

The policy should say that AI output is a draft until a person has checked it. The Audit and Beyond article puts it this way: treat AI-generated content as a first draft, and apply the same scrutiny as you would to a junior team member’s work. ICAEW’s PCRT guidance says members remain ultimately accountable for any work produced, regardless of whether AI has been involved, and that due care can be applied by performing due diligence on the output from the AI tool using a risk-based approach.

The policy should settle four points.

  • Who may approve output on each type of work. The FRC’s March 2026 guidance, written for audit firms, says those who review AI output need the appropriate competence to identify potential deficiencies.
  • What the check consists of. The same FRC guidance says review may involve forming an independent view to compare the output against, and names professional scepticism and awareness of automation bias.
  • What is recorded. The PCRT guidance suggests members consider maintaining an audit trail, including details of the AI tools used, in case of later queries. The Audit and Beyond article says to document inputs and outputs on file.
  • What staff must disclose. The PCRT guidance says a staff member who uses an AI tool in their work should disclose this to senior colleagues.

The policy also needs one sentence saying that the partner who signs remains responsible. A policy cannot delegate that responsibility to a tool.

Confidentiality and data handling

The Code of Ethics position

Paragraph 114.0 A1 of the Code says a safe approach is to assume that all unpublished information about a client’s affairs is confidential. Paragraph R114.1(d) requires reasonable steps to ensure that personnel under the accountant’s control, and individuals from whom advice and assistance are obtained, comply with the duty of confidentiality, and 114.1 A1 says protecting confidentiality covers the collection, use, transfer, storage or retention, dissemination and lawful destruction of the information. Under R114.2(d) the duty applies even where information has become publicly available. The exception in R114.3 includes disclosure authorised by the client.

On tools, ICAEW’s generative AI ethics guidance says respecting confidentiality means not loading confidential information into public generative AI tools, because there is limited visibility and control over who the information is shared with, how it is secured and how long it is retained. The PCRT guidance says input of client data into publicly available AI tools is likely to constitute a breach of client confidentiality unless the client has consented, and that data put into publicly available models should be anonymised and generic.

Data protection

Where a tool provider processes personal data on the practice’s behalf, Article 28(1) of the UK GDPR says the controller shall use only processors providing sufficient guarantees to implement appropriate technical and organisational measures. The policy can require that this is checked before a tool is approved. The ICO lists processing involving innovative technologies, including AI, among the types that call for a data protection impact assessment when combined with another high-risk criterion, and says that if in any doubt it would always recommend that you do one. Checked against the ICO page on 21 November 2026; the ICO notes that its guidance is under review following the Data (Use and Access) Act. Our guide to client data, UK GDPR and AI tools takes these points in full.

Training, and reviewing the policy

Paragraph R113.2 of the Code requires an accountant to take reasonable steps to ensure that those working under their authority have appropriate training and supervision, and the Code lists receiving training on identifying bias among the actions that might mitigate it. The FRC’s March 2026 guidance, again for audit firms, says staff education is an ongoing activity rather than something relevant only at the launch of a tool. For a practice, that points to training when a tool is approved and again when it changes, covering what the tool is for, its known weaknesses, and the confidentiality rules.

The Audit and Beyond article recommends a culture where staff feel safe to try new tools but are clear on what is acceptable. A policy that people can read in a few minutes is more likely to be followed than one they have to look up.

We recommend that the policy carry a review date and be revisited when a tool is added or changes materially. That is our recommendation rather than a requirement. It fits the FRC’s own statement that its AI guidance does not represent a static set of expectations and that the field is moving quickly, and the advice of an ICAEW technical adviser, quoted in its engagement letter update, that engagement letters should be reviewed regularly and that the adoption of new digital tools should prompt an update to the terms agreed with the client.

Questions practices ask

How long should an AI use policy be?

No source sets a length. An ICAEW Audit and Beyond article by a partner in a smaller firm recommends a simple policy that is concise and focuses on the key principles, so that staff are clear what is expected.

Does the policy have to ban public AI tools?

No source says so. ICAEW’s guidance is that confidential client information should not be loaded into public generative AI tools, and that client data put into public models should be anonymised and generic. Whether to allow public tools for other work is a decision for the practice.

Do staff have to tell senior colleagues when they use AI?

ICAEW’s PCRT guidance says they should, and suggests members consider keeping an audit trail of the tools used. A policy is a natural place to make that a rule.

Is the policy the same as what we tell clients?

No. The policy governs staff. What clients are told belongs in the engagement letter and any client-facing statement, and ICAEW’s PCRT guidance notes that clients can be directed to a data handling or AI usage policy on the firm’s website.

Sources