What is Cowork?Use CasesBlogServicesAI Readiness QuizFAQBook a Call

Blog · Regulatory

Using AI in a licensed financial services business — what ASIC expects

A practical 2026 guide for AFSL holders, authorised representatives and Responsible Managers. Report 798, REP 775, s912A general obligations, DDO, SOA quality, record keeping and breach reporting — what the regulator has actually said and what it means for your practice.

The short answer

Can Australian AFSL holders use AI?

Yes. ASIC has not restricted AI use in licensed financial services. The regulator's consistent position across Report 798, REP 775 and public statements from Chair Joe Longo in 2024 and 2025 is that the existing obligations apply in full — s912A general obligations, the FASEA Code of Ethics, the best interests duty, DDO under RG 274, record keeping under s912G and breach reporting under s912DAA. The licensee owns the outcome. The AI tool is a drafting assistant, not a defence.

ASIC's AI focus: what the regulator has actually said

Between 2023 and late 2025, ASIC moved from scoping AI use to publishing structured expectations. The most relevant documents for financial advisers:

  • Report 798 — Beware the gap: Governance arrangements in the face of AI innovation (2024). Based on a review of 23 licensees, REP 798 found governance frameworks lagging the pace of AI deployment. Its core finding — that boards often lacked visibility of where AI was being used across their business — applies as readily to advice licensees as to the insurers surveyed.
  • REP 775 — Discussion on algorithmic decision-making in consumer credit and financial services (2023). The paper set ASIC's early expectation that automated decisions affecting consumers must be explainable, fair and auditable. Advisers using AI to triage clients, prioritise review books or pre-filter product shortlists sit squarely inside its themes.
  • Surveillance activities (2024–2025). ASIC's enhanced surveillance programme has flagged AI governance as an ongoing focus. Chair Joe Longo's September 2024 address to the Committee for Economic Development of Australia described AI as a priority supervisory area and signalled that ASIC will "act where consumer harm emerges" without waiting for new legislation.

The practical read: ASIC is not drafting a bespoke AI rulebook. It is applying the Corporations Act 2001 and the FASEA Code as written, to a new class of tool.

AFSL obligations and AI: s912A applied to drafting tools

Section 912A of the Corporations Act sets the general obligations every AFSL holder operates under. Three limbs matter most for AI deployment:

  • Efficiently, honestly and fairly (s912A(1)(a)). AI tools used to draft SOAs, review letters or client correspondence must produce outputs that are honest and fair — no fabricated figures, no hallucinated product features, no overstated benefits. The licensee's processes must catch these before the client sees anything.
  • Adequate resources and risk management (s912A(1)(d) and (h)). AI use needs a governance wrapper: documented acceptable-use policy, a register of tools in use, training, and a review process for outputs. REP 798 made clear that "the vendor is managing it" is not an adequate risk framework.
  • Competence (s912A(1)(e) and (f)). Advisers and support staff using AI tools need enough competence to recognise when an output is wrong. That is a training obligation, not a procurement one.

FASEA Code of Ethics and the Financial Advisers Register

The FASEA Code applies to every relevant provider on the Financial Advisers Register. Two standards carry most of the AI weight:

  • Standard 5 — informed consent. Clients must understand the advice and the basis for it. If AI materially shaped the drafting, Standard 5 points towards disclosing the use of AI tools in the Financial Services Guide and engagement letter. Many Australian licensees are updating these documents through the 2026 review cycle.
  • Standard 7 — fees for no service risk. If AI is used to generate SOAs or review letters that are not genuinely personalised — where the adviser is not meaningfully engaged with the client's circumstances — the fee-for-no-service risk re-emerges. ASIC's enforcement history on FFNS is long; AI efficiency is not a shield.

The licensed adviser's name on the SOA is the point of professional responsibility. That does not move when an AI tool enters the workflow.

Design and Distribution Obligations (RG 274) and AI-assisted product selection

DDO requires issuers and distributors to take reasonable steps to ensure financial products reach consumers in their target market. For advisers, that lands most visibly in product shortlisting. If AI tools are used to pre-filter products against a client's circumstances, three guardrails apply:

  • The Target Market Determination still governs. AI cannot route a client outside their TMD without a documented reasonable basis.
  • The adviser's analysis sits above the AI output. The RG 274 obligation is on the distributor, not the model.
  • Complaints and feedback loops feeding DDO monitoring (s994E) still need to be captured, whether the shortlist came from a human or a tool.

Statement of Advice quality: reviewer responsibility remains

The SOA is the single most scrutinised document in Australian advice. Where AI drafts it, the licensed adviser's review is what makes it compliant. In practice:

  • Every numerical figure in the SOA needs to be verified against source documents — fact-find, platform statements, PDSs. AI tools can transcribe figures correctly and can also transpose them.
  • The "reasons why" section needs to reflect the adviser's actual reasoning, not generic strategy language. If the AI drafted it from a thin brief, the reviewer rewrites it.
  • Alternatives considered and risks disclosed must be genuinely tailored. Template-sounding language is a well-known audit flag.

Reviewer responsibility is not a tick-box. It is the step where the s961B best interests duty is discharged.

Record keeping under s912G: prompts and outputs on the file

Section 912G requires AFSL holders to retain records sufficient to demonstrate compliance with their obligations. For AI-assisted work, that extends to the workflow itself:

  • The prompt used to generate the draft (or a reference to the saved template)
  • The raw AI output before editing
  • The adviser's edits and the final signed document
  • Any tool configuration (model, system prompt, data source) that materially affected the output

Seven years is the standard retention period. This mirrors the audit trail most practices already keep for paraplanner-drafted work — it is a format change, not a new obligation.

Breach reporting under s912DAA

The reportable situations regime applies to AI-related breaches the same way it applies to any other cause. A materially misleading SOA, a fee-for-no-service outcome, a privacy incident involving client financial data — each attracts the 30-day reporting clock from the point the licensee has reasonable grounds to believe a reportable situation has arisen. The presence of an AI tool is neither a defence nor an automatic aggravator. What ASIC will examine is the governance around the tool: did the licensee have reasonable steps in place?

Practical framework: a five-step compliance checklist

For AFSL holders rolling AI into their advice workflow in 2026, a workable ASIC-aligned framework:

  1. Register and policy. Maintain a register of AI tools in use across the practice. Publish an acceptable-use policy covering what data can be entered, who can use which tool, and what review is required before any output reaches a client.
  2. Data handling. Confirm your Privacy Policy covers AI processing. On paid Claude Cowork plans Anthropic does not train on your data; verify equivalent positions for every tool you use. Redact TFNs, Medicare numbers and full account numbers before any upload.
  3. Reviewer workflow. Every client-facing AI-drafted document is reviewed and signed off by the authorised representative. The Responsible Manager or compliance team samples outputs regularly.
  4. Record keeping. Save the prompt, the output, the adviser's edits and the final document to the client file. Seven-year retention.
  5. Breach pathway. Build AI-caused errors into your existing reportable situations process. Train staff to escalate early; the 30-day clock is unforgiving.

None of this is new regulatory ground. It is the s912A general obligations, applied methodically to a tool the regulator expects licensees to govern.

FAQ

Common questions on ASIC and AI

Does ASIC allow AI in financial advice?

Yes. ASIC has not banned AI use in licensed financial services. Its position, consistent across Report 798, REP 775 and public statements from Chair Joe Longo, is that existing obligations — the general obligations under s912A, the FASEA Code of Ethics, DDO under RG 274 and the best interests duty — apply in full. The licensee remains responsible for outputs whether or not AI was involved.

Do I need to tell clients I used AI to draft their SOA?

ASIC has not mandated a specific AI disclosure to retail clients, but FASEA Standard 5 requires informed consent and Standard 2 requires honesty. If AI materially shaped the advice (not just formatting), good practice is to note the use of AI drafting tools in your FSG and engagement letter. Many AFSL holders are updating these in the 2026 review cycle.

What records should I keep when AI is used to draft advice?

Section 912G requires records sufficient to demonstrate compliance. For AI-assisted work, keep the prompt, the AI output, the adviser's edits and the final signed document on the client file. Seven years is the standard retention period. This mirrors the paraplanner audit trail most practices already maintain.

Is an AI-generated SOA a breach of the best interests duty?

No, provided the licensed adviser retains professional judgment and the SOA genuinely reflects the client's circumstances and goals. The best interests duty under s961B is a duty of the adviser, not the tool. An SOA that is AI-drafted but adviser-reviewed, personalised and accurate meets the standard. One that is rubber-stamped without genuine review would not.

When is an AI issue a reportable situation under s912DAA?

If AI contributes to a breach of a core obligation — a materially misleading SOA, a fee-for-no-service outcome, or a privacy breach — the reportable situations regime applies the same way it would for any other cause. The 30-day clock starts when the licensee has reasonable grounds to believe a reportable situation has arisen.

Does ASIC's Report 798 apply to financial advisers?

REP 798 focuses on general and life insurers, but ASIC has been explicit that its themes — governance, explainability, consumer outcomes, fair treatment, oversight — apply across licensees. Advisers using AI for SOA drafting, portfolio commentary or client triage should read REP 798 as a signal of regulatory expectation across the broader AFSL population.

Get your AFSL practice AI-ready

Claude Cowork training for Australian financial advisers — ASIC-aligned workflows, licensed adviser sign-off retained, SOA drafting compressed by 70%.

See Training OptionsBook a Discovery Call

From A$249 per person · Team (up to 5) A$549 · Implementation from A$3,500