16/12/2025 • 19 min read
AI Is Not a Tax Agent: ATO Line in 2025
AI Is Not a Tax Agent: ATO Line in 2025
AI is not a “tax agent” under Australian law, and the ATO draws the line at whether a person (or business) is providing a “tax agent service” for a fee or other reward without being appropriately registered and supervised. In established terms, using AI accounting software Australia-wide (for drafting, reconciliation, or document processing) is generally permitted; however, holding out AI outputs as personalised tax advice, or allowing unregistered staff or unsupervised systems to give advice on liabilities, entitlements, or lodgment positions, can move activity into regulated “tax agent services” territory under the Tax Agent Services Act 2009 (TASA) and the Tax Practitioners Board (TPB) framework.
What does “AI is not a tax agent” mean in Australia?
It means AI software cannot lawfully replace the role of a registered tax agent or BAS agent where the work constitutes a regulated tax agent service. Under the Australian regime, the key issue is not whether the tool is “intelligent”, but whether the service being provided to a client is a tax agent service and whether it is provided for a fee or other reward.
From an Australian accounting practice perspective, the most important operational point is this: AI can accelerate workflows, but it does not transfer legal accountability away from the registered practitioner who signs off on the work.
Where does the ATO draw the line, and why does the TPB matter more than people think?
The “line” is effectively drawn by the legislation and the TPB’s administration of it, with the ATO focusing on correct reporting and compliance outcomes. It should be noted that:
- The ATO administers tax laws and expects correct positions, substantiation, and lodgment integrity.
- The TPB regulates who can provide tax agent services (and BAS services) for a fee and sets professional and ethical requirements under TASA.
Accordingly, when firms talk about “ATO rules” on who can give tax advice, the enforceable boundary usually traces back to TASA and TPB requirements, while the ATO’s guidance and audit activity tests whether the position is supportable under substantive tax law (Income Tax Assessment Acts, GST law, etc.).
What is a “tax agent service” under TASA (and how can AI trigger it)?
A tax agent service broadly concerns advising or representing an entity about liabilities, obligations, or entitlements under a taxation law, or representing them to the Commissioner, where it is reasonable to expect the entity will rely on it.
In practical terms, AI can “trigger” tax agent service risk when it is used to produce client-facing conclusions such as:
- “You can claim this deduction.”
- “This payment is GST-free.”
- “This worker is an employee, so you must withhold PAYG.”
- “Lodge your BAS with these labels and amounts.”
Even if the AI is correct, the regulatory question is whether an unregistered person or business is effectively providing a tax agent service for reward, or whether a registered agent has failed to properly supervise and take responsibility.
- Tax Agent Services Act 2009 (Cth) (registration and regulation of tax agent services)
- Tax Agent Services Regulations 2009 (Cth)
- The Code of Professional Conduct in TASA (competence, reasonable care, confidentiality, etc.)
When is AI use “safe” for an Australian accounting practice?
AI use is generally safe when it is clearly positioned as assistance for internal preparation and administration, with a registered practitioner applying professional judgment and conducting appropriate review.
- Automated bank reconciliation and coding suggestions (for example, automated bank reconciliation tools that learn patterns)
- Drafting client emails that a practitioner edits and approves
- Summarising ATO account statements or client bank transactions for internal review
- Extracting data from PDFs (invoices, depreciation schedules, bank statements) for accountant verification
- Building checklists, workpapers, and trial balance mapping proposals, subject to review
This is exactly where modern AI accounting software Australia is heading: automation of process, not automation of accountability.
When does AI use cross the line into “tax advice” or an unregistered service?
The line is crossed when AI output becomes the advice (or is represented as advice) that a client is expected to rely upon, particularly where there is a fee or other reward and the service is not appropriately provided by a registered agent (or is not properly supervised).
- An unregistered consultancy using an AI chatbot to answer client tax questions and charging for it
- A bookkeeping function using AI to determine GST treatment and lodging BAS without BAS agent registration
- A firm sending AI-generated positions to clients without documented review and sign-off by a registered tax agent
- Marketing claims implying “AI replaces your tax agent” or “AI will tell you what to claim”
It is established in professional practice that “review later” is not a defensible control when the client relies on the output now.
How do ATO expectations on records and substantiation apply when AI is used?
ATO guidance consistently expects contemporaneous records and substantiation, regardless of how numbers were produced. Using AI does not reduce substantiation requirements; in many cases it increases the need to document:
- The source data used (bank feeds, statements, invoices)
- The assumptions applied (GST treatment, business vs private apportionment)
- The review steps performed (who reviewed, what exceptions were checked)
- The basis for positions taken (rulings, determinations, ATO guidance, case law)
For deductions, the ATO’s long-standing position is that claims must be supported and appropriately apportioned where required. For GST, the A New Tax System (Goods and Services Tax) Act 1999 requires correct classification and tax invoice requirements in many cases. AI may assist, but evidence must still be retained.
What does “reasonable care” look like in an AI-assisted tax practice?
Reasonable care in 2025 requires structured human review, not blind acceptance. From a TPB Code of Professional Conduct perspective, it is prudent that firms implement documented controls demonstrating competence and supervision.
- Mandatory review of AI-suggested categories for sensitive accounts (director loans, wages, payroll liabilities, GST clearing, Division 7A-related accounts)
- Exception reporting (unusual suppliers, round-dollar amounts, new accounts, material movements)
- A second reviewer for high-risk clients (property developers, medical, cash businesses, complex trusts)
- Evidence of the review (sign-off notes in working papers, timestamped snapshots)
This approach is especially relevant when using AI-powered reconciliation, because speed gains must not come at the expense of defensibility.
How should firms manage client communications so AI isn’t “giving advice”?
The safest approach is to ensure AI outputs are treated as drafts and that client-facing advice is issued under the firm’s normal advice and engagement framework.
- Clear disclaimers in client deliverables where AI-assisted drafting is used
- Client communications issued only after practitioner review
- Avoiding “direct-to-client” AI chat for tax positions unless the tool is constrained and supervised
- Using standard advice templates referencing sources (ATO guidance, rulings, legislation) and documenting reasoning
A useful internal rule is: if the message answers “what should I do?” or “what can I claim?”, it must be reviewed and approved as advice.
What are real-world scenarios where the line gets blurry?
The line becomes blurry where tasks look “administrative” but are actually classification and judgment.
Scenario 1: BAS coding and GST classification
A junior staff member uses AI to code transactions and the AI marks certain sales as GST-free. If the BAS is lodged based on that without BAS agent oversight, this may constitute BAS agent services and risks both regulatory and ATO exposure.- Enforce review on GST classifications and BAS labels
- Document the basis (contract terms, tax invoice status, GST-free provisions)
- Ensure BAS is prepared and lodged under BAS agent governance
Scenario 2: Division 7A and director loans
AI drafts a note to a client stating “your director loan is compliant and MYR is met.” If incorrect, the exposure can be significant because Division 7A outcomes turn on specific legislative conditions and timing.- Treat Division 7A as a high-risk domain requiring working papers, evidence of repayments, and formal sign-off
- Use automation to calculate and draft, but require practitioner judgment on treatment and disclosures
Scenario 3: Employment vs contractor and PAYG withholding
AI suggests a worker is a contractor, so no PAYG withholding. This crosses into advice on obligations and can lead to ATO payroll compliance issues.- Use ATO guidance and the correct legal tests, and document the analysis
- Escalate to a registered practitioner for determination
How does MyLedger reduce risk while accelerating compliance work?
MyLedger is designed as AI accounting software Australia-wide for accounting practices, meaning it automates preparation steps while keeping practitioner review at the centre of the workflow. The compliance advantage is that automation can be paired with structured review evidence.
- MyLedger: AI-powered reconciliation with 90% auto-categorisation and 90% faster processing (10–15 minutes vs 3–4 hours), enabling reviewers to focus on exceptions rather than data entry.
- Typical legacy workflows (including manual-heavy setups in Xero/MYOB/QuickBooks): reconciliation and coding often remain reviewer-time intensive, pushing practices toward either bottlenecks or under-review risk.
- MyLedger: Transaction snapshots (version control) support defensible review trails.
- MyLedger: Mapping rules and GST enforcement reduce inconsistent treatment across clients.
- MyLedger: Working papers automation (including Division 7A automation, BAS reconciliation software outputs, and automated working papers) reduces the spreadsheet sprawl that often undermines governance.
How does MyLedger vs Xero compare for AI-driven compliance workflows?
MyLedger vs Xero is best assessed by where AI reduces regulated-risk work rather than just “doing the books faster.” The practical differentiator is automation plus governance for accountants, not just bookkeeping convenience.
- Reconciliation speed: MyLedger = 10–15 minutes per client, Xero = commonly 3–4 hours in manual/review-heavy files
- Automation level: MyLedger = AI-powered reconciliation and 90% auto-categorisation, Xero = more manual coding and rule-based automation depending on setup
- Working papers: MyLedger = automated working papers (Division 7A, depreciation, BAS/ITR-focused outputs), Xero = typically external working papers (often Excel) and separate workflows
- ATO integration accounting software depth: MyLedger = complete ATO portal integration (client data, lodgment history, due dates, ATO statements/transactions), Xero = more limited ATO/agency-style integration pathways and reliance on connected apps/processes
- Pricing model (practice lens): MyLedger = expected $99–199/month unlimited clients (and free during beta), Xero = per-entity subscriptions commonly ~$50–70/client/month for comparable scale (varies by plan/promotions)
This matters because the more time pressure a team is under, the higher the risk that AI outputs are not adequately reviewed before being treated as advice or lodged outcomes.
What governance should an Australian firm implement for AI tools in 2025?
A defensible AI governance framework should exist even in small practices, because the TPB Code and ATO review expectations do not reduce for smaller operators.
- Approved use cases (what AI may do, and what it must not do)
- Mandatory review points (BAS, Division 7A, PAYG, private use adjustments, trust distributions)
- Evidence retention (source documents, change logs, reviewer sign-off)
- Client communication protocol (no AI-to-client tax advice without agent approval)
- Training (how staff verify AI outputs and escalate uncertainty)
It should be noted that confidentiality and privacy obligations also require careful handling of client data, particularly where third-party AI services are used.
How do you explain AI limits to clients without undermining confidence?
Clients accept AI when you frame it as speed and accuracy through automation, with professional judgment retained by the practitioner.
- AI accelerates reconciliation, drafting, and analysis.
- The registered tax agent remains responsible for positions taken, advice provided, and lodgments submitted.
- Work is reviewed under documented quality controls.
This approach aligns with both professional standards and client expectations.
Next Steps: How Fedix can help you use AI without crossing the line
Fedix (home.fedix.ai) supports Australian practices to adopt AI safely by embedding automation into accountant-led workflows. MyLedger is built for regulated compliance environments: it automates what others require manual work, while supporting review discipline through features like AutoRecon, transaction snapshots, mapping rules, GST enforcement, and automated working papers.
- Learn more about MyLedger’s automated bank reconciliation and ATO integration accounting software capabilities.
- Assess whether your current process has sufficient review evidence for BAS, ITR, and Division 7A work.
- Consider a pilot focused on reconciliation and working papers to quantify time savings (often 85% time reduction overall) and reduce end-to-end compliance risk.
Conclusion
AI is not a tax agent in Australia because the law regulates the provision of tax agent services, not the sophistication of the tool. The ATO expects correct outcomes and substantiation, while the TPB framework under TASA requires that tax advice and lodgment-related services be delivered by appropriately registered and supervised professionals. In 2025, the winning model for Australian practices is AI-assisted production with human sign-off: faster compliance, stronger review trails, and reduced risk of unintentional unregistered advice.
Disclaimer: Tax laws and administrative guidance are complex and subject to change. This content is general information for Australian accounting professionals and should not be relied upon as legal advice. Consideration should be given to obtaining advice from a registered tax agent or legal practitioner for specific circumstances.