ATO on AI 2025: ChatGPT & Automated Tax Advice

The ATO’s current stance is that AI tools such as ChatGPT can be used to support tax administration and professional work, but they do not replace a taxpayer’s or agent’s legal obligations to keep proper records, take reasonable care, and lodge correct statements; if AI-generated content is wrong, the taxpayer/agent remains responsible. In practice, the ATO expects AI outputs to be treated as unverified drafts, with appropriately qualified review, evidence, and governance—particularly where advice influences positions taken under the tax law.

What does the ATO actually say about using AI like ChatGPT for tax?

The ATO’s public messaging consistently emphasises that you must rely on authoritative sources and keep evidence for your tax positions, even when technology is used to assist. This reflects the ATO’s administration of Australia’s self-assessment system and its focus on data integrity, substantiation, and correct reporting.

From a practical compliance perspective, the ATO position can be summarised as follows:

• AI can assist with drafting, summarising and workflow support, but it is not authoritative.
• ATO guidance, legislation, and (where relevant) case law remain the primary sources for tax positions.
• “Garbage in, garbage out” applies: AI output depends on the prompt, data provided, and model limitations.
• Responsibility does not shift to the tool provider; it stays with the taxpayer and agent.

Australian practice implication: if a client says “ChatGPT told me I could claim it,” that is not a substantiation strategy. Evidence, nexus to income, and correct characterisation under the relevant provisions remain mandatory.

Is ChatGPT “tax advice” in the ATO’s view?

In ATO compliance terms, ChatGPT output is not a substitute for professional advice or binding guidance and should be treated as general information unless verified and tailored to the client facts. The ATO’s frameworks for advice and reliance are grounded in:

• The hierarchy of legal authority: legislation first, then relevant case law, then ATO public rulings/determinations and guidance.
• The ATO’s published advice products (for example, public rulings, practical compliance guidance, and online guidance), which are designed to communicate the Commissioner’s view.

Practice reality: AI outputs frequently omit conditions, exceptions, integrity rules, dates of effect, and definitions. That is precisely where tax risk sits.

Why does the ATO care about AI-generated content?

The ATO cares because AI increases the volume and speed of content creation, which can amplify incorrect claims, poor substantiation, and inconsistent reporting. The ATO’s compliance posture is heavily data-driven and focused on:

• Correctness of the statement lodged (BAS/IAS/ITR)
• Substantiation and record keeping
• Governance over tax outcomes for larger and more complex taxpayers
• Agent behaviour and “reasonable care” processes

From an Australian accounting practice perspective, the risk is not “using AI”; the risk is using AI without controls and then lodging.

What legal obligations still apply if AI drafted the tax position?

The legal obligations remain unchanged when AI is used. Key obligations commonly engaged include:

• Record keeping requirements under the tax law (including keeping records that explain transactions and support claims).
• The self-assessment framework: taxpayers must lodge correct returns based on their circumstances.
• Penalty and interest regimes, which turn on behaviour (for example, failure to take reasonable care, recklessness, or intentional disregard).
• Tax agent professional standards and duties when advice is provided and returns are prepared.

Relevant anchor points for practitioners (high level):

• Tax administration and record-keeping obligations administered under the Taxation Administration Act 1953 (for example, substantiation/record retention concepts and administrative penalty framework).
• Income tax rules in the Income Tax Assessment Act 1997 and, where relevant, the Income Tax Assessment Act 1936.
• Public rulings and guidance where the ATO sets out its interpretation (for example, practical guidance and rulings relevant to the issue being advised on).

Professional point: an AI prompt and response is rarely “contemporaneous evidence” of the underlying transaction. The ATO typically expects source documents (invoices, contracts, logbooks, bank evidence), not an AI narrative.

Does using ChatGPT increase penalty risk with the ATO?

Using ChatGPT does not automatically increase penalty risk, but relying on unverified AI output can. The ATO’s penalty analysis generally turns on behaviour and care taken.

From a risk lens for Australian practices:

• If AI output is used as a draft and the agent verifies facts, law and evidence, risk is controlled.
• If AI output is used as the basis for positions without verification, the likelihood of incorrect statements rises, which can trigger adjustments, penalties and interest.

Typical “AI-caused” failure modes seen in practice:

• Incorrect thresholds and rates (e.g., GST, super, withholding) because the model is out-of-date.
• Misapplication of residency tests, PSI rules, or CGT main residence exceptions due to missing facts.
• Hallucinated citations (invented cases, rulings, or legislative sections).
• Overconfident language that hides uncertainty.

ATO-aligned control principle: demonstrate reasonable care through documented review and substantiation, not through the sophistication of the tool.

How should an Australian accounting practice safely use AI under ATO expectations?

A compliant approach is to treat AI as a productivity tool inside a governance framework. The minimum control set should include:

• Source-of-truth rule:
• Verification checklist before lodgment:
• Prompt hygiene and confidentiality:
• Version control and audit trail:
• Escalation rules:

What are real-world examples of “ATO on AI” issues in practice?

AI risk becomes tangible at lodgment and review time. The following scenarios reflect common practice pitfalls.

Example 1: Work-related expenses justified by “ChatGPT said it’s deductible”

• Evidence (invoice/receipt), and
• A clear business connection, and
• Apportionment where there is mixed use.

• Draft a client questionnaire and a deduction workpaper.
• Generate a checklist of substantiation documents to request.

• Produces generic “yes you can claim” answers without apportionment and evidence requirements.

Example 2: GST/BAS coding errors from auto-categorisation

• Flag anomalies (e.g., “GST-free”, “input-taxed”, “no ABN withholding risk indicators”).
• Generate exception lists for review.

• Automatically applies GST to items that are GST-free or input-taxed, creating BAS errors that the ATO can detect via data matching.

Example 3: Division 7A “advice” generated by ChatGPT

• Draft minutes, checklists, and repayment schedule explanations for the client.
• Summarise the ATO’s published approach to common fact patterns (then verify).

• Misses the need for a complying loan agreement by the required time, misstates benchmark rates, or ignores payments/offsets classification.

How does the ATO’s stance interact with ATO Online, practice systems, and “ATO integration accounting software”?

The ATO’s operational direction is toward better data quality, integration, and traceability. “ATO integration accounting software” can support compliance by reducing transcription error and maintaining audit trails.

From a practice workflow standpoint:

• ATO-linked data (statements, transactions, lodgment history, due dates) reduces reliance on manual copying.
• Automated reconciliation that ties bank data to working papers improves substantiation readiness.
• AI should sit behind controls so outputs are reviewable and evidence-backed.

This is where “AI accounting software Australia” differs from generic AI chat tools: workflow AI that is embedded in reconciliation, working papers, and ATO-integrated processes is inherently more controllable than open-ended text generation.

How does MyLedger compare to generic AI tools (like ChatGPT) for ATO-aligned work?

ChatGPT is a general language model, not an accounting system of record. MyLedger (by Fedix) is designed for Australian accounting practices, where the compliance outcome depends on reconciled data, workpapers, and traceability.

Key practice differences (no tables; feature-by-feature):

• Primary purpose: MyLedger = accounting automation and compliance workflow, ChatGPT = text generation
• Evidence base: MyLedger = bank data + transaction detail + working papers, ChatGPT = whatever text you provide
• ATO integration: MyLedger = direct ATO portal integration (client data, lodgment history, due dates, ATO statements/transactions), ChatGPT = none
• Reconciliation speed: MyLedger = 10–15 minutes per client (about 90% faster than 3–4 hours manual-heavy processes), ChatGPT = cannot reconcile bank data as a system workflow
• Working papers: MyLedger = automated working papers (including BAS reconciliation and Division 7A automation), ChatGPT = drafts narratives only
• Audit trail: MyLedger = snapshots/versioning and structured outputs, ChatGPT = conversational logs that are not accounting records

• Use ChatGPT-style drafting for client communications, then anchor all numbers and positions in MyLedger’s reconciled dataset and working papers.

What should your firm policy say about using AI for tax and BAS work?

A firm AI policy should be explicit that AI is an assistant, not an authority, and must be paired with verification and evidence. A robust Australian policy typically includes:

• Acceptable use:
• Prohibited use without approvals:
• Mandatory steps:

How Fedix can help (Next Steps)

Fedix built MyLedger to operationalise ATO-aligned compliance workflows, not just generate text. If your practice is assessing “ChatGPT and automated advice” risk, the most effective control is moving from narrative-based work to data-backed automation.

1. Standardise your reconciliation workflow with MyLedger AutoRecon to reduce BAS and year-end error rates and cut reconciliation time to 10–15 minutes per client.
2. Use MyLedger’s ATO integration to pull ATO statements/transactions and due dates directly, reducing manual handling.
3. Implement automated working papers (including Division 7A automation and BAS reconciliation) so positions are supported by structured evidence, not AI narratives.

Learn more at home.fedix.ai and evaluate whether MyLedger is the right “Xero alternative” for an Australian practice focused on automation, governance, and ATO-ready workpapers.

Conclusion

The ATO’s stance on AI is not “don’t use it”; it is that AI must not replace evidence, legal authority, and reasonable care. For Australian accounting practices, the safest approach is to use AI for drafting and efficiency while anchoring all outcomes in reconciled data, working papers, and ATO-aligned governance—precisely the gap that practice-grade platforms like MyLedger are designed to close.

Frequently Asked Questions

Q: Is the ATO okay with taxpayers using ChatGPT for tax questions?

Q: Can I rely on ChatGPT output to reduce ATO penalties?

Q: What should accountants do when a client says “ChatGPT said I can claim it”?

Q: Does the ATO treat AI-generated records as adequate substantiation?

Q: How can my firm use AI while staying ATO-compliant?

Disclaimer: This content is general information for Australian accounting professionals as of December 2025 and does not constitute legal or tax advice. Tax laws and ATO guidance can change, and outcomes depend on specific facts. Consider obtaining advice from a suitably qualified professional for your circumstances.