Accounting AI software is already inside Australian accounting firms, extracting invoice data, classifying transactions, drafting tax notes, summarising audit working papers and drafting client correspondence. Every one of those uses touches a specific obligation under APES 110, the TPB Code of Professional Conduct, ATO record-keeping rules, the Privacy Act 1988 or, from July 2026, AML/CTF law. The work to make AI safe is mostly governance, supervision and records.
Key takeaways
- Accounting firms face a stack of regulators: TPB, APESB (APES 110), ATO, OAIC, ASIC where advice is given, AUSTRAC for designated services, plus CA ANZ, CPA Australia and IPA standards.
- Each AI use case triggers different obligations, so map them one by one instead of treating "AI in the practice" as a single thing.
- Confidentiality (APES 110 Subsection 114) is the most common issue: client data must not go to AI tools that retain or train on it.
- A registered tax agent remains responsible for AI-assisted work. Supervision models, review thresholds and retained working papers make that defensible.
- Most firms need a pre-deployment, live and review control cycle more than they need a better model.
Which regulators look at AI in an accounting firm?
For partners and practice managers in firms covering business services, tax, audit and advisory, AI exposure is unusually layered. Healthcare and aged care face concentrated regulators. Financial services face APRA and ASIC. Accounting firms face a profession-specific ethics code, a profession-specific registration regime and the standards of three member bodies, with no single front door.
Treat these as a stack, because the same AI workflow can trigger several obligations at once.
| Body | What it covers for AI | |
|---|---|---|
| TPB (Tax Practitioners Board) | Tax Agent Services Act 2009, Code of Professional Conduct, 2024 Code Determination | Competent provision of tax agent services, supervisory arrangements, keeping clients informed, breach reporting |
| APESB / APES 110 | Code of Ethics for Professional Accountants (including Independence Standards) | Integrity, professional competence and due care, confidentiality, professional behaviour |
| ATO | Record-keeping rules, lodgement integrity, GST, PAYG, FBT, STP | Substantiation of AI-assisted positions, retention of AI working papers, the lodger remains responsible |
| ASIC | Corporations Act, AFSL obligations | Where the firm holds an AFSL or gives financial advice, AI in advice processes must meet licensee obligations |
| OAIC | Privacy Act 1988, Australian Privacy Principles, Notifiable Data Breaches scheme | APP 1 governance, APP 6 use and disclosure, APP 8 cross-border, APP 11 security, NDB notification |
| AUSTRAC | AML/CTF Act, extended to accountants providing designated services from 1 July 2026 | AI-assisted customer due diligence, suspicious matter reporting and screening must remain auditable |
| CA ANZ / CPA / IPA | Member standards adopting APES 110 plus body-specific guidance | Member discipline, body-issued AI guidance, CPD obligations |
Two points stand out. The TPB Code was strengthened in 2024 with obligations on honest conduct, supervisory arrangements and breach reporting that apply directly to AI use. And OAIC Notifiable Data Breaches reports regularly list legal, accounting and management services among the top reporting sectors, so the privacy layer is a live risk.
What can accounting AI software do, and which obligations does each use trigger?
The common mistake is treating AI in the practice as one thing. Each use case sits inside a different set of obligations.
| Use case | Primary obligations triggered + minimum control | |
|---|---|---|
| Invoice and receipt data extraction | Source documents to GL, often via Xero/MYOB or document AI | APES 110 confidentiality, APP 11 security, APP 8 if processed offshore. Control: data residency clause, no training on client data, privacy impact assessment on file |
| Bank reconciliation and categorisation AI | ML suggesting account codes, matching feeds to invoices | APES 110 competence and due care, TPB Code competent provision. Control: documented review threshold, accuracy sampling, exception handling |
| Tax return draft generation | AI drafting schedules, deductions, narrative for lodgement | TPB Code, ATO record-keeping, APES 110 competence. Control: practitioner review of every position, retained working paper showing AI output and human override |
| Audit working paper summarisation | LLM summarising controls testing, walkthroughs, evidence | ASA 230 audit documentation, APES 110 competence. Control: source-linked summary, reviewer signs off original evidence, AI summary is not the audit record |
| Client-facing communication (advice drafts, query triage) | Drafting emails, answering FAQs, generating advice templates | APES 110 integrity, TPB Code, ACL misleading conduct, liability for AI representations. Control: accountant approval before send, disclosure where material, scope limits on any client-facing assistant |
| Practice management (workflow, time, billing) | Workflow prediction, time capture, narrative generation, billing | Privacy Act (staff data), Fair Work where it touches performance, APES 320 and APES 325. Control: quality system documents AI tools used, risk register includes AI risks |
The same vendor can sit in three rows at once. A general-purpose LLM with access to your document management system is a confidentiality risk, a competence risk (juniors can produce work above their skill level without realising) and a record-keeping risk (prompts and outputs may not be kept).
Can AI catch GST compliance issues?
Yes, as a first-pass reviewer. AI can flag transactions whose GST coding looks inconsistent with the supplier, description or past treatment, find missing tax invoices above the ATO threshold, and highlight unusual input tax credit claims before a BAS is lodged. The accountant still decides the treatment. Edge cases such as going concern supplies, mixed supplies and margin schemes are exactly where general models give the typical answer and the typical answer is wrong, so set flags for human review instead of auto-correction.
Can AI triage client queries and summarise ATO guidance?
Yes, with a qualified accountant in the loop. A workable pattern: AI classifies incoming client queries by type and urgency, drafts a response from the firm's own precedents and current ATO guidance with citations, and routes it to the responsible accountant, who approves or edits before anything goes to the client. For summarising ATO rulings and guidance, require the AI to link every statement to its source so the reviewer can check it quickly. The approval step is what keeps this within APES 110 and the TPB Code.
If you are choosing models, our AI model selection guide and AWS, Azure and Google Cloud AI services in Australia cover the residency and processing-location questions to answer first.
How does APES 110 apply to AI confidentiality and competence?
APES 110 is the Code of Ethics issued by the Accounting Professional and Ethical Standards Board (APESB). Four fundamental principles bite hardest on AI:
Confidentiality (Subsection 114). Members must not disclose confidential information acquired through professional and business relationships outside the firm without proper authority. Sending client data to a third-party LLM that retains, logs or trains on it is a disclosure outside the firm, and the most common APES 110 issue we see. The mitigations are well known: enterprise terms with no-training clauses, processing in Australia or AI running inside your own environment, and minimising data in prompts. They only work with a documented policy staff follow.
Professional competence and due care (Subsection 113). Members must maintain the knowledge and skill needed for competent professional service. The AI is a tool and the practitioner owns the output. Two failure modes: a senior relies on AI output they cannot verify (over-reliance), or a junior produces senior-looking work that is signed off without proper review (false competence).
Integrity (Subsection 111). Be straightforward and honest. The live question is disclosure to clients. APES 110 has no specific AI disclosure rule, but TPB obligations to keep clients informed, and member body guidance, push firms toward disclosure where AI is materially involved. A practical test: would the client be surprised, and would that surprise make them question the work?
Professional behaviour (Subsection 115). Comply with laws and avoid conduct that discredits the profession. A leaked client file through an unauthorised AI tool, or an AI-generated tax position that causes a penalty, becomes a professional behaviour issue.
APES 320 (Quality Management for Firms that provide Non-Assurance Services) and APES 325 (Risk Management for Firms) sit alongside. Both expect the firm's quality and risk systems to cover how it controls technology, including AI. If your APES 320 documentation does not mention AI tools, it is out of date.
How does the TPB Code apply to AI in tax work?
The Tax Practitioners Board registers tax agents and BAS agents under the Tax Agent Services Act 2009. The Code of Professional Conduct was strengthened in 2024, with obligations covering honest and ethical conduct, taking on only work the practitioner is competent to do, supervisory arrangements, keeping clients informed of matters affecting their decisions, and reporting breaches.
The TPB's position is that AI does not change registrant obligations: a registered tax agent is responsible for the work, however it was produced.
- AI produces outputDraft return, schedule, classification, position narrative
- Practitioner reviewRegistered agent verifies the work meets the competence obligation
- Supervisory arrangementJunior using AI is supervised under the firm's documented model
- Client disclosureMaterial AI involvement disclosed in engagement letter or work papers
- Working paper retainedAI prompt, output, and human override retained per ATO record-keeping rules
- Breach reportingIf an AI error causes a Code breach, follow the TPB reporting obligation
Supervision is where firms most often fall short. A graduate with a capable LLM can produce work that looks senior. Without a documented supervisory model that addresses AI-assisted work, the firm cannot show the TPB that the work was properly supervised. Firms also need a process for recognising when an AI-related error amounts to a Code breach, and making sure the responsible registrant knows their reporting obligation.
What does ATO record-keeping mean for AI-generated outputs?
The ATO's position on AI is the same as for any tool: the lodger is responsible for the lodgement. What changes is the record-keeping question.
Missed exceptions. AI trained on general tax content returns the typical answer. Australian tax law is full of exceptions, integrity measures and timing rules where the typical answer is wrong: Division 7A, CGT concessions, FBT exemptions, GST going concerns, residency edge cases. Mitigate with a documented review checklist the practitioner runs against AI output before lodgement.
Working paper retention. The ATO generally requires records to be kept for five years, longer in some cases. When AI sits between source documents and the final return, the prompt, the output and the human review and override all form part of the working paper trail. Without them, the firm cannot show how a queried position was reached.
Substantiation. "The AI said so" is not a defensible answer. The defensible answer is: the AI produced a draft, the practitioner reviewed it against these sources and this reasoning, and the position was kept for these documented reasons.
Operating AI agents in production and the AI agent staffing gap cover the operational layer this needs.
How do the Privacy Act, NDB scheme and AML/CTF rules apply?
Most accounting firms are APP entities, and the Privacy (Tax File Number) Rule applies to any firm that handles tax file numbers. Four APPs do most of the work:
- APP 1, open and transparent management: the privacy policy must accurately describe how personal information is handled, including by AI systems. From 10 December 2026, APP 1.7 also requires disclosure of certain automated decisions, covered in our guide to Privacy Act automated decision making.
- APP 6, use and disclosure: information collected for tax preparation cannot be used for a secondary purpose, such as training a vendor's model, without consent or another lawful basis.
- APP 8, cross-border disclosure: if AI processing happens outside Australia, take reasonable steps to ensure the overseas recipient does not breach the APPs, or rely on an exception. Our GDPR vs Privacy Act guide covers the mechanics.
- APP 11, security: take reasonable steps to protect personal information. Client tax data in a personal chatbot account does not meet APP 11. A procured service with proper contractual and technical controls, or AI running on your own infrastructure in Australia, can.
The Notifiable Data Breaches scheme is the consequence layer. If an AI exposure leaks personal information likely to cause serious harm, the firm must notify the OAIC and affected individuals. Firms hold TFNs, bank details, salaries and dependant information, so the harm threshold is easily met.
AML/CTF now reaches accountants directly. Under the AML/CTF Amendment Act 2024, accountants providing designated services became reporting entities from 1 July 2026. AI-assisted customer due diligence, screening and suspicious matter reporting must remain auditable and explainable, and the reporting entity stays accountable. Our guide to AML/CTF tranche 2 for accounting firms goes into detail, and the AI security checklist is the operational companion.
What controls does an accounting firm need for AI?
Controls cluster into a pre-deployment, live and review cycle, whether you are switching on a Xero AI feature, a document automation tool, an LLM for drafting or a workflow agent.
- Pre-deploymentVendor and use case assessmentDocument the use case against APES 110, TPB Code, ATO record-keeping. Vendor data residency, training clauses, sub-processor list. APES 320 and APES 325 updates.
- Pre-deploymentPrivacy and security reviewAPP 1 policy update, APP 8 cross-border assessment, APP 11 security controls, privacy impact assessment where personal info is in scope. NDB response plan updated.
- Pre-deploymentSupervisory and disclosure modelDocumented supervisory model for AI-assisted work. Engagement letter and client disclosure language. Breach reporting flow to TPB.
- LivePractitioner review and working paper retentionDocumented review threshold for AI output by use case. Retention of prompt, output, and override in the working paper trail. Sampling regime for quality.
- LiveException handling and incident responseDefined exception categories, escalation paths, and incident triggers for NDB and TPB reporting. Staff trained on what to escalate and when.
- ReviewQuarterly AI register reviewAPES 320 quality review of AI tools in use. APES 325 risk register refresh. Vendor reassessment for material changes (model updates, sub-processor changes).
- ReviewAnnual independent reviewIndependent review of the firm's AI governance against APES 110, TPB Code, and Privacy Act obligations. Outcomes feed the next year's quality plan.
This is the minimum that lets a firm show it has met its obligations. It is mostly documentation and consistent practice. Our automation business case template and AI vendor selection questions support funding and procurement.
Which obligation is the binding constraint for your AI deployment?
More than one obligation usually applies, but one tends to decide whether a deployment can go live and on what terms.
What does the AI actually do with client information?
- Processes client confidential information offshore without contractual data residency controlBLOCKED. APP 8 + APES 110 confidentiality. Fix residency or change vendor before proceeding.
- Produces a tax position, advice, or audit conclusionTPB Code + APES 110 competence are binding. Practitioner review and supervisory model are non-negotiable controls.
- Sends output to a client without practitioner reviewAPES 110 integrity + ACL misleading conduct + liability for AI representations. Require an accountant to approve before send.
- Processes personal information that would trigger NDB if leaked (TFNs, banking, dependants)APP 11 + NDB scheme are binding. Security controls and incident plan are non-negotiable.
- Touches a designated service under AML/CTFAUSTRAC is the binding regulator. AI output must remain explainable and auditable for due diligence and reporting.
- Only processes the firm's own internal data with no client personal informationPrivacy Act partially relevant; APES 320 quality management and APES 325 risk management still apply.
On client communication: in Moffatt v Air Canada (2024 BCCRT 149), Air Canada was held liable for information its chatbot gave a customer. Australian Consumer Law has its own misleading conduct framework and the principle reads across. A firm is responsible for representations its AI makes to clients, by email, chat or any other channel. Our guide to ACCC consumer guarantees and AI covers the ACL angle, and financial services AI compliance under APRA and ASIC applies the same stack thinking to firms holding an AFSL.
What should partners check before expanding AI?
Work through these twelve questions. Any you cannot answer with documented evidence is your next piece of work.
- Which AI tools are in use across the firm today, including informal use by individual staff?
- For each tool, what client information passes through it, and where is it processed?
- Does the vendor contractually commit to no training on your data and to specified data residency?
- Has the privacy policy been updated for AI processing of personal information?
- Is there a documented supervisory model for AI-assisted work that satisfies the TPB Code?
- Is there a documented review threshold by use case (for example, tax positions reviewed line by line, internal drafts sampled)?
- Are prompts, outputs and human overrides retained in the working paper trail in line with ATO record-keeping?
- Has APES 320 quality management documentation been updated to include AI tools?
- Does the APES 325 risk register include AI risks (hallucination, data leakage, vendor change, model drift)?
- Is there a documented approach to disclosing material AI involvement in engagement letters or work papers?
- Does the NDB response plan cover AI incidents, including who contacts the OAIC, the TPB and affected clients?
- Is there a quarterly review of the AI register, vendor assessments and incident log?
The firms that get this right have a documented system that would survive a TPB review, an OAIC investigation or a member body referral. The number of tools has nothing to do with it.
How Solve8 can help
Our consultants work with professional services firms to map their current AI footprint against the TPB, APES 110, ATO, Privacy Act and AML/CTF obligations, then build the controls and workflows on your infrastructure in Australia, with an accountant approving before anything reaches a client or the ATO. We built Carbonly.ai, our own compliance reporting product, using the same review-first approach. See our AI strategy services or talk to us.
Sources: APESB APES 110 Code of Ethics, APES 320, APES 325; Tax Practitioners Board guidance and the 2024 Code of Professional Conduct Determination; ATO record-keeping guidance; OAIC Notifiable Data Breaches reports; AUSTRAC guidance and the AML/CTF Amendment Act 2024; Moffatt v Air Canada 2024 BCCRT 149.
