ICAS and AI: What Scottish Accountancy Practices Must Do Before Deploying Client-Facing AI Tools | Insights

This series has focused on Scottish law firms. But the compliance architecture is sector-agnostic — and accountancy practices regulated by the Institute of Chartered Accountants of Scotland (ICAS) face identical data protection obligations, an equally stringent professional ethical framework, and client data categories that are, in many respects, more uniformly sensitive than those held by solicitors. Every tax return contains a National Insurance number. Every set of financial statements reveals the commercial health of a business. Every piece of HMRC correspondence is confidential by statute.

If your accountancy practice tolerates Shadow AI — staff pasting client tax data into ChatGPT, associates uploading financial statements to Gemini, trainees using free AI summarisers on HMRC correspondence — the exposure is the same as for a law firm. The ICO does not distinguish between regulated professions when calculating penalties. The DUAA 2025 applies equally. The maximum fine is £17.5 million or 4% of global turnover regardless of sector.

For the ICO enforcement risks facing AI-adopting professional services firms in Scotland, read our parent article. This piece maps the ICAS-specific ethical obligations that constrain AI adoption — and demonstrates that the sovereign AI architecture satisfying the Law Society of Scotland’s requirements satisfies ICAS’s requirements identically.

The Five Fundamental Principles: Technology-Updated From 1 January 2025

The ICAS Code of Ethics, substantively based on the International Ethics Standards Board for Accountants (IESBA) Code, establishes five fundamental principles that all Chartered Accountants must observe. From 1 January 2025, ICAS adopted revised IESBA provisions specifically addressing the impact of technology — including AI — on professional conduct.

The five principles, each with direct AI implications:

1. Integrity. Straightforward and honest in all professional relationships. Using AI to generate client-facing work product without disclosure — or without verifying accuracy — is a failure of integrity. The 2025 revisions emphasise that accountability rests with the human professional regardless of the tool used.

2. Objectivity. Professional judgement without compromise. The revised Code explicitly highlights the risk that over-reliance on technology could impair objectivity. If a Chartered Accountant relies on an AI-generated tax analysis without independent verification, and the output contains a hallucinated tax relief that does not exist, the accountant’s objectivity has been compromised by the tool.

3. Professional competence and due care. Attain and maintain knowledge at the level required to ensure competent service. The 2025 revisions include new provisions on “acting with sufficient expertise” in the context of technology. A Chartered Accountant must understand the AI tool’s capabilities, limitations, and failure modes before relying on its output. ICAS’s own research — the “Generative AI and Professional Judgement in Accounting” report, published in 2026 — found that 52% of respondents raised concerns about client data privacy and confidentiality when using generative AI.

4. Confidentiality. Respect the confidentiality of information acquired through professional relationships. The 2025 Code includes a new formal definition of “confidential information”: any information, data, or other material in whatever form or medium that is not publicly available. The revised provisions require professional accountants to safeguard confidential information throughout its entire data lifecycle — acquisition, storage, processing, and disposal. Transmitting client tax data to ChatGPT violates the confidentiality principle at the processing stage. The information has left the accountant’s control and entered a third-party system with no contractual confidentiality protection.

5. Professional behaviour. Comply with relevant laws and regulations. Avoid conduct that discredits the profession. Using public AI tools on client data in contravention of UK GDPR, without a DPIA, without processor governance, and without client consent, is a failure of professional behaviour that could result in ICAS disciplinary proceedings.

The Data Sensitivity Profile: Why Accountancy Data Is Uniformly High-Risk

Scottish law firms hold a wide spectrum of data sensitivity — some matters involve minimal personal data, others involve the most sensitive information imaginable. Accountancy practices hold data that is uniformly high-risk:

National Insurance numbers — a unique personal identifier directly tied to HMRC records, benefits, and state pension. Present in virtually every client engagement.
Tax returns and self-assessment data — income, capital gains, property holdings, pension contributions, charitable donations. Comprehensive financial profiles of individuals and businesses.
HMRC correspondence — statutory communications containing tax calculations, penalty notices, investigation letters. Confidential by statute and by professional obligation.
Financial statements — profit and loss, balance sheet, cash flow. Commercially sensitive to the point where premature disclosure could affect market valuations, lending decisions, or acquisition negotiations.
Payroll data — employee names, addresses, salaries, tax codes, National Insurance contributions. Personal data at scale.

Every category is personal data under UK GDPR. Several categories constitute financial data that, if disclosed, could cause substantial harm to the data subject. The ICO’s penalty calculation methodology weights the sensitivity of data categories when assessing seriousness at Step 1. Accountancy client data starts the assessment high.

The Compliance Architecture Is Identical

The ICAS Code of Ethics and the Law Society of Scotland’s conduct rules arrive at the same destination through different statutory instruments. Both require confidentiality of client data. Both require professional competence. Both require compliance with data protection law. Both require human accountability for technology-assisted output.

The sovereign AI architecture that satisfies the Law Society’s requirements satisfies ICAS’s requirements without modification:

Confidentiality (ICAS Principle 4 / Law Society Rule B1.6): Zero-egress architecture. Client data never leaves the firm’s contracted server. Sealed network environment with no outbound data path. No third-party AI vendor receives any data.

Professional competence (ICAS Principle 3 / Law Society Rule B1.10): Open-weight models with documented capabilities and known limitations. Model manifest delivered at handover. The Chartered Accountant understands what the tool is, how it works, and what it cannot do.

Integrity and objectivity (ICAS Principles 1–2): All AI output is advisory. Professional sign-off by a qualified Chartered Accountant is mandatory. The firm’s governance framework requires human review of every AI-assisted work product before it is relied upon or communicated to clients.

Professional behaviour (ICAS Principle 5): UK GDPR compliance demonstrated via DPIA, Article 30 records, executed DPA with the infrastructure provider, and the Technical Configuration Letter confirming Private AI Ltd’s Zero Standing Privileges architecture and Default Operating State. DUAA 2025 obligations met via producible technical artefacts.

Technology-specific confidentiality safeguards (2025 Code revisions): Confidential information safeguarded throughout its entire data lifecycle — from prompt submission through inference, embedding, storage, and eventual deletion — entirely within the firm’s zero-egress perimeter.

What ICAS-Regulated Practices Must Do Now

The regulatory trajectory is clear. ICAS adopted technology-specific ethics revisions from 1 January 2025. Its “Generative AI and Professional Judgement” research programme published findings in 2026 emphasising that human judgement cannot be replaced by AI and that data privacy concerns are shared by the majority of the profession. The ICO’s enforcement record — a sevenfold increase in fines during 2025 — applies to accountancy practices with the same force as to law firms.

If your practice has not yet addressed AI data processing, the preparation steps are:

Audit for Shadow AI using the methodology described in our earlier article — DNS logs, browser extensions, email forensics, and staff interviews apply to accountancy practices identically
Complete a DPIA covering all AI data processing, mapping each risk to a specific technical mitigation
Ensure your processor map is complete — every AI vendor identified, every DPA executed, every international transfer verified. Or deploy sovereign AI with two directly contracted processors and no AI-layer vendor
Update your terms of engagement to address AI usage, in line with the ICAS confidentiality principle’s requirement to safeguard information throughout its data lifecycle
Brief your partners on the ICO’s 2025 enforcement record and the DUAA 2025 enforcement powers now in force

The compliance infrastructure is the same. The regulatory obligations are the same. The risk is the same. The solution is the same.

Talk to our team to understand how a sovereign AI deployment maps to the ICAS Code of Ethics and UK GDPR obligations specific to accountancy practices.

Private AI Ltd (SC881079)