ATO on AI 2025: ChatGPT & Automated Advice

The ATO’s stance on AI tools (including ChatGPT and other generative AI) is that they may be used to assist with tax and accounting work, but they do not replace the taxpayer’s or agent’s legal responsibility for accuracy, substantiation, privacy, and proper record-keeping. In practice, the ATO expects Australian accountants to treat AI output as unverified draft material, apply professional judgement, maintain evidence for claims, and comply with agent and taxpayer obligations under Australia’s tax laws and the ATO’s guidance on record keeping, data security, and correct reporting.

What is the ATO’s current position on ChatGPT and AI-generated tax advice?

The ATO’s practical position is that AI can be a productivity tool, but AI output is not a primary source of tax law and cannot replace authoritative references such as ATO guidance, public rulings, and legislation. This aligns with the ATO’s broader messaging that taxpayers and agents must “get it right,” maintain records, and be able to substantiate positions.

From an Australian accounting practice perspective, the ATO’s stance can be summarised as:

• AI-generated content is not ATO advice and is not binding on the Commissioner.
• Responsibility remains with:
• Claims must be supported by contemporaneous records and evidence, consistent with ATO record-keeping expectations.
• Privacy, confidentiality, and security must be managed when client data is entered into AI tools.

Which ATO sources should you treat as authoritative (not ChatGPT)?

The ATO treats the following as primary and reliable sources (and they are the sources you should cite internally when validating AI outputs):

• ATO website guidance (including guidance by topic and by form schedule instructions)
• ATO public rulings and determinations (for example, Taxation Rulings “TR”, Practical Compliance Guidelines “PCG”, Tax Determinations “TD”)
• ATO Law Administration Practice Statements (PS LA series) where relevant to administration
• Relevant legislation, including (commonly):

It should be noted that an AI model may quote or paraphrase these sources inaccurately, out of date, or without crucial qualifiers.

Is using ChatGPT for tax work “allowed” in Australia?

Yes, using ChatGPT is generally permissible, but it must be governed by professional standards, privacy requirements, and robust verification controls. The ATO does not police “tool choice” in isolation; it examines outcomes (correct reporting, substantiation, governance) and behaviours (reasonable care, honesty, proper records, correct disclosures).

In practical terms, an Australian practice may use generative AI for:

• Drafting client communications (with review)
• Summarising ATO web guidance and rulings (with citation checking)
• Producing first-draft checklists and working paper narratives
• Drafting internal training notes (with validation)

An Australian practice should avoid using generative AI as the final decision-maker for:

• Complex technical positions (for example, Division 7A, trust distributions, residency, CGT events)
• Interpretive conclusions without checking current rulings/legislation
• Any advice where facts are incomplete, disputed, or evolving

What legal and professional obligations still apply when AI is used?

The legal obligations do not change because AI is involved; AI is merely a tool in the workflow. The key compliance lens is: could you demonstrate reasonable care, accuracy, substantiation, and proper record-keeping if reviewed by the ATO?

What does “reasonable care” mean when AI is involved?

Reasonable care, in ATO terms, generally requires competent processes to ensure correct reporting and supportable positions. When AI is used, reasonable care should include:

• Validating AI output against:
• Confirming the client’s factual matrix (AI cannot do this reliably)
• Documenting assumptions, exclusions, and materiality thresholds in working papers
• Retaining substantiation (invoices, contracts, logbooks, bank evidence, payroll records, etc.)

How do record-keeping expectations interact with AI drafting?

ATO guidance on record keeping is outcome-focused: records must be kept that explain transactions and support claims. If AI is used to generate narratives or calculations, the practice should retain:

• Source documents (primary evidence)
• The calculation methodology (and any spreadsheets/working papers)
• The review trail (who reviewed, what changed, why)
• Version control for key workpapers (particularly year-end and BAS)

This is where purpose-built accounting automation platforms can materially reduce risk by producing consistent, repeatable working papers and audit trails, rather than ad hoc prompts and pasted outputs.

What are the biggest ATO risks when relying on ChatGPT or automated advice?

The ATO risks are not theoretical; they map directly to common failure points that lead to amended assessments, penalties, and poor audit outcomes.

Key risks include:

• Hallucinations and false citations
• Out-of-date answers
• Missing Australian context
• Privacy and confidentiality breaches
• Overconfidence bias

How should an Australian accounting practice use AI safely (and in a way the ATO would expect)?

A safe approach is to treat generative AI as a junior drafting assistant and enforce a documented verification process.

What governance controls should be implemented?

A robust governance framework typically includes:

• An “AI Use Policy” approved by partners/directors covering:
• A verification checklist for any AI-assisted tax position:

What is a practical verification workflow for AI-assisted advice?

1. Use AI to draft a summary or client explanation in plain language.
2. Identify the relevant topic area and locate current ATO web guidance.
3. Identify relevant public rulings/determinations (TR/TD/PCG) if interpretive guidance is required.
4. Confirm legislative references (ITAA 1997/1936, GST Act, TAA 1953 as applicable).
5. Apply the client facts and document the reasoning in your working papers.
6. Perform partner/manager review for high-risk positions.
7. Keep the final advice with references and substantiation.

What real-world scenarios show the difference between “AI assistance” and “ATO-defensible advice”?

AI can be beneficial, but only if the practice designs the workflow so the final output is evidence-based and sourced.

Scenario 1: BAS and GST coding using AI

A common use case is transaction categorisation and reconciliation support.

• AI-appropriate:
• Not AI-appropriate as final:

In practice, ATO scrutiny of GST reporting focuses on correct classification and substantiation. AI should accelerate processing, not replace GST technical analysis.

Scenario 2: Division 7A and “quick answers” from ChatGPT

Division 7A is a frequent audit risk area and is highly fact-dependent.

• AI-appropriate:
• Not AI-appropriate as final:

ATO expectations here are strict: working papers must align with ATO guidance and the legislative framework in ITAA 1936 (Division 7A). Any AI output must be reconciled to those requirements.

Scenario 3: Deduction substantiation and “work-related” claims

• AI-appropriate:
• Not AI-appropriate as final:

The ATO’s compliance approach routinely focuses on substantiation and whether claims are properly evidenced.

How does the ATO view automation in accounting compared to “automated advice”?

The ATO generally supports digitalisation that improves accuracy, timeliness, and record integrity, but “automated advice” that produces legal conclusions without controls is a risk. The key distinction is:

• Automation of processing (reconciliation, classification, report generation) can improve compliance outcomes when it creates consistent records and audit trails.
• Automation of interpretive tax advice is risky unless it is tightly governed, reviewed, and anchored to current authoritative sources.

From a practice standpoint, the lowest-risk and highest-ROI AI is workflow automation that reduces manual handling while improving consistency and traceability.

What tools and controls best align with ATO expectations in 2025?

ATO-facing defensibility improves when your systems produce consistent working papers, clear reconciliation support, and verifiable source data.

For many firms, the best-fit tooling approach is:

• Use generative AI for:
• Use accounting automation platforms for:

MyLedger (Fedix) vs general-purpose AI chat tools: what is the compliance advantage?

General-purpose AI chat tools are not accounting systems and typically do not produce accounting-grade audit trails.

• Primary purpose: MyLedger = accounting automation and working papers, ChatGPT-style tools = text generation
• Reconciliation outcome: MyLedger = 90% faster reconciliation (10–15 minutes vs 3–4 hours), ChatGPT-style tools = cannot reconcile bank data end-to-end
• Evidence trail: MyLedger = transaction-level workflow, snapshots/version control, and exportable reports, ChatGPT-style tools = prompt/output text with limited governance
• ATO integration: MyLedger = complete ATO portal integration (including ATO statement and transaction import), ChatGPT-style tools = no ATO portal integration
• Australian compliance workflow: MyLedger = BAS summaries, GST enforcement, Division 7A working papers automation, ChatGPT-style tools = generic drafting with no embedded compliance controls

This matters because the ATO’s review activity focuses on how figures were derived and whether claims are substantiated—not whether the narrative is well written.

How should practices explain AI use to clients (without increasing liability)?

Client communications should clearly position AI as an internal productivity tool under professional supervision, not a source of truth.

A defensible client disclosure approach is:

• State that:
• Emphasise that:

Next Steps: How Fedix can help Australian practices use AI safely

If the objective is to adopt AI in a way that aligns with ATO expectations, the priority should be workflow automation that strengthens records and reduces manual handling risk.

Fedix, through MyLedger, is designed specifically for Australian accounting practices to operationalise AI safely:

• Automate bank reconciliation with AutoRecon (10–15 minutes per client vs 3–4 hours; approximately 90% faster).
• Generate consistent working papers (including Division 7A automation and BAS reconciliation) to support ATO-defensible outcomes.
• Use complete ATO portal integration to import ATO statements and transactions and reduce re-keying risk.
• Standardise practice defaults across clients to improve consistency and review quality.

Learn more at home.fedix.ai and assess where automation can replace manual steps without outsourcing professional judgement.

Conclusion

The ATO’s stance on ChatGPT and automated advice is effectively “use with caution”: AI can assist productivity, but it is not authoritative and does not reduce legal responsibilities for correct lodgments, substantiation, and record keeping. Australian practices that adopt AI safely will treat it as drafting support, validate outputs against ATO guidance and legislation, and prioritise systems that strengthen audit trails, working papers, and reconciliation integrity. MyLedger by Fedix is positioned to deliver that compliance-aligned automation, replacing manual steps that commonly introduce errors.

Frequently Asked Questions

Q: Does the ATO accept ChatGPT output as evidence or authority?

Q: Can a registered tax agent use AI to draft advice and client emails?

Q: What is the biggest compliance risk of using generative AI in tax?

Q: Should AI be used for Division 7A calculations and MYR schedules?

Q: How can a practice show “reasonable care” if AI was used?

Disclaimer: This content is general information for Australian accounting professionals as of December 2025 and does not constitute legal or tax advice. Tax law and ATO guidance are complex and subject to change; specific matters should be reviewed against current ATO publications, applicable rulings, and legislation, and tailored to the client’s circumstances.