Home / AI for accountants / AI in audit
AI in auditing
AI in audit
AI in audit means using machine learning, generative AI and, increasingly, agentic tools inside audit engagements, from risk assessment and journals testing to reading contracts and summarising board minutes.
This guide goes deeper than our overview of AI for accountants: it sets out what the UK regulator has actually published, what that means for documentation and professional scepticism, and where data analytics already changes the shape of testing. It is not a substitute for your firm's methodology or the auditing standards themselves.
Understand what the FRC expects of firms using AI tools, how to document a tool proportionately, and how to keep the auditor's judgement visible on file.
On 26 June 2025 the FRC published its first guidance on AI in audit, including principles for proportionate documentation of AI tools; it is not prescriptive and introduces no new regulatory requirements. FRC: guidance on the uses of AI in audit ↗
FRC guidance of 30 March 2026 on generative and agentic AI in audit codifies good practice; the human auditor remains accountable and firms should consider ISQM (UK) 1. FRC: Generative and Agentic AI guidance (March 2026) ↗
New to the topic? Read the background on artificial intelligence.
AI in auditing
what has the FRC actually published on AI in audit?
Two pieces of guidance: an illustrative example with documentation principles in June 2025, and guidance on generative and agentic AI in March 2026.
The FRC's June 2025 guidance has a two-part structure: an illustrative example of one way AI can be used in an audit, and principles intended to support proportionate and robust documentation of tools that use AI. Its definition of AI is broad, covering traditional machine learning and deep learning, including generative AI.
On 30 March 2026 the FRC followed with Generative and Agentic AI Guidance. The FRC describes it as the second guidance it has issued on AI in audit and states that it is not a response to identified quality deficiencies; it codifies good practice seen in firms and provides a conceptual foundation for future work.
Both documents sit together on the FRC's AI in Audit library page, alongside a factsheet. Read them in the original before changing methodology: summaries, including this one, lose nuance.
Neither publication rewrites the auditing standards. The practical question for a firm is how its existing quality management and engagement procedures absorb a new kind of tool.
June 2025: illustrative example and documentation principles.
March 2026: generative and agentic AI risks and mitigations.
Neither changes who is accountable.
AI in auditing
who is accountable when an AI tool does audit work?
The firm and the engagement partner, exactly as before.
The FRC's factsheet states that the firm and engagement partner retain full responsibility for audit quality, in accordance with ISQM (UK) 1 and ISA (UK) 220 respectively.
The March 2026 announcement is equally direct: regulatory accountability for deploying AI tools and for the quality of audit outputs is unchanged, and the human auditor is always accountable. Firms are encouraged to consider how their use of AI tools fits within wider quality management obligations under ISQM (UK) 1.
For AI in audit, that turns a technology decision into a quality management decision. Someone at the firm has to own the tool, its approved uses, and how staff are expected to challenge what it produces.
It also means an engagement team cannot treat a vendor's assurances as a substitute for its own work. The FRC's 2025 guidance specifically clarifies how expectations apply where a tool is obtained from a third party.
AI in auditing
how should a firm document an AI tool without over-documenting?
Proportionately: enough to show why the output can be relied on in that use, not a technical dossier for every prompt.
The 2025 guidance describes its documentation expectations as balanced and proportionate, intended to prevent over-documentation. It also acknowledges that the appropriate level of explainability varies with context and usage.
In practice that suggests documenting the tool once at firm level (what it is, what it is approved for, how it was tested) and then documenting at engagement level only what the team relied on and how they checked it.
The FRC says its documentation guidance reflects the UK government's five AI principles. Firms already using those principles in their AI policy can cross-reference rather than duplicate.
Treat the illustrative example as an example. The FRC presents it as one potential way AI can be used, and firms will need to apply the principles to their own tools and use cases.
| Check | What to establish |
|---|---|
| 1 | Firm-level record: purpose, approval, testing. |
| 2 | Engagement-level record: reliance and checks. |
| 3 | Scale detail to risk and explainability. |
AI in auditing
what happens to professional scepticism with AI in audit?
It moves to the output: the auditor has to obtain appropriate confidence in what the tool produced before relying on it.
The March 2026 guidance frames the question as obtaining appropriate confidence in the quality of a tool's output, and sets out principles for exercising professional judgement in doing so.
The factsheet groups mitigations into four categories: system design and development, certification, staff education and governance, and human-in-the-loop review and oversight. It notes that the extent of activity in each category is a matter of professional judgement.
The FRC's two illustrative use cases are summarising board minutes and using an AI tool to review contracts for revenue recognition testing. Both are tasks where a fluent, plausible output can hide an omission, which is exactly where scepticism has to be deliberate.
ICAEW's Head of Data Analytics and Tech has noted that many auditors are cautious about generative AI because of output consistency and risks such as hallucinations, and that good AI needs good data (ICAEW). That caution is a feature of scepticism, not resistance to change.
AI in auditing
does AI in audit mean testing every transaction instead of sampling?
Sometimes, for some populations, but that comes from data analytics as much as from generative AI, and it does not remove judgement.
ICAEW, discussing the FRC's 2026 Annual Review of Audit Quality, notes that analytics allows auditors to go beyond testing relatively small samples, and that some tools can test 100% of populations in a client's financial records, potentially more frequently during the period (ICAEW).
The same article points to analytics for risk assessment, journals and revenue testing, and back-testing management's impairment models for patterns of bias under ISA 540.
Full-population testing changes the question rather than answering it. Every exception still needs to be followed up, the completeness and integrity of the data extracted from the client's systems still matter, and a clean run over bad data is still bad evidence.
Generative AI tends to help with unstructured material such as contracts and minutes, while deterministic analytics handles structured ledgers. Being clear which is which keeps the documentation honest.
Validate data completeness before full-population runs.
Plan how exceptions will be investigated.
Distinguish analytics from generative AI on file.
AI in auditing
how should an audit team build skills for AI in audit?
Train on the firm's own approved tools and real engagement tasks, with review and documentation built into the exercise.
ICAEW keeps a running artificial intelligence resource page covering the FRC guidance, CCAB ethics material and free generative AI e-learning for members. It is a reasonable reading list before designing internal training.
Training that works for auditors is task-based: take a real, anonymised board pack or contract set, run it through the approved tool, and then do the review the FRC expects, writing up what was checked.
predeploy runs AI training for accountants on exactly these workflows, alongside policy and tool-settings work. We train people; we do not train models on client data, and we do not provide audit methodology or accreditation.
If your firm is still choosing tools, the AI tools for accountants guide covers evaluation criteria, and AI training for finance teams covers the in-house side of the same skills.
Use real, anonymised engagement material.
Practise the review, not just the prompt.
Align training with firm methodology.
Make the decision concrete
What does the current task cost in time?
Choose a task connected to ai in audit and enter its current workload. This gives you a baseline to discuss before buying training or commissioning a build.
A baseline is useful even if you decide to keep the process manual. Measure the work that matters, then compare quality and review effort as well as time.
Assess the workflow before changing it, or plan business AI training if the gap is how people use existing tools.
Your current workload
Estimate how much time the existing task takes before discussing ai in audit.
16 hours per week
People × minutes × weekly tasks ÷ 60. A workload baseline from your inputs, not a prediction of AI time savings. Include review and support time when comparing a proposed workflow.
A useful AI investment leaves your team able to do the work and judge the result.
Continue with the right question
Where should you go next?
Use the overview for context, then follow the pages that match your delivery decision.
See our ai training service or read the ai for accountants overview.
FAQ
Questions, answered
Evidence and further reading
Which sources support the guidance?
Source links support the specific background described here. The practical checklists and illustrative examples are predeploy’s guidance.
On 26 June 2025 the FRC published its first guidance on AI in audit, including principles for proportionate documentation of AI tools; it is not prescriptive and introduces no new regulatory requirements. FRC: guidance on the uses of AI in audit ↗
FRC guidance of 30 March 2026 on generative and agentic AI in audit codifies good practice; the human auditor remains accountable and firms should consider ISQM (UK) 1. FRC: Generative and Agentic AI guidance (March 2026) ↗
Firm and engagement partner retain full responsibility under ISQM (UK) 1 and ISA (UK) 220; mitigations grouped as system design, certification, staff education and governance, and human-in-the-loop review. FRC: Generative and Agentic AI Guidance factsheet ↗
FRC page holding the 2025 illustrative example and documentation guidance and the 2026 generative and agentic AI guidance and factsheet. FRC: AI in Audit library page ↗
Analytics lets auditors go beyond small samples; some tools can test 100% of populations; supports ISA 540 challenge of impairment models. ICAEW: Why data analytics is crucial in improving audit quality ↗
ICAEW's Ian Pay notes auditor caution about generative AI output consistency and hallucinations, and that good AI needs good data. ICAEW: Data analytics for auditors: five essentials ↗
ICAEW hub listing FRC generative and agentic AI guidance, CCAB ethics guidance, HMRC AI expectations and free GenAI e-learning. ICAEW: Artificial intelligence resources ↗
Background reading on the broader field of artificial intelligence. Wikipedia: artificial intelligence ↗
Choose your next step
Explore ai for accountants
Start with a topic overview, or go directly to the question your team needs to answer.
AI for accountants
Uses, tools, governance and training for practices and finance teams.
Start with the decision
More questions in this topic
