Built to keep you compliant — and to prove it

Everything the platform produces — tax analysis, general tax information and modelling, savings estimates, AI Assistant answers, reports and draft communications — is general information generated by automated systems, not advice. As the registered practitioner you review, verify, and exercise your own professional judgement on every output before it reaches a client. This is written into the Firm & Tax-Agent Agreement you accept on application — it is the basis of the relationship, not fine print.

The AI Assistant runs behind layered guardrails: input is screened, the topic is held to Australian tax, output is filtered, and a "general information, not advice" disclaimer is injected at the platform level — the model cannot suppress it. When a conversation moves past general information toward personalised advice, the assistant declines and points back to you, the registered agent. See the AI disclaimer and safety measures.

Before an AI request leaves our trust boundary, a de-identification layer detects structured identifiers (such as TFN-formatted values and ABNs) and replaces them with non-identifying placeholders, which are restored on the way back. For each request we keep an internal transparency ledger recording which categories of identifiers were replaced — not the values themselves. This is detailed in our Privacy Policy. De-identification is high-recall, not perfect, so it sits alongside — not instead of — the guardrails and your own review.

Where an answer is grounded in source material, the assistant cites the ATO page or the strategy it drew on, so you and your client can trace a position back to its authority rather than taking the model's word for it. How we produce the underlying numbers is documented in our open methodology.

When your team acts on AI-assisted output, the platform records a timestamped acknowledgement that the practitioner understood it was general information and would verify it independently before advising — at onboarding, and at the point of acting on an answer or exporting a report. The exact wording shown is versioned and hashed, so the record proves what was acknowledged, not just that a box was ticked. No client tax file numbers or personal identifiers are stored in that log.

From your firm settings, an admin can export a TPB Compliance Defence Pack for any date range: a branded PDF plus a CSV bundling your AI interactions (with the categories of identifiers that were masked), your document-access history, the platform strategy endorsements current in the period, and your team's verification acknowledgements. It is the artefact you produce if you are ever asked to show your oversight of AI-assisted work. The pack contains no raw client identifiers, and it is a record of platform activity — not an endorsement, certification, or assurance by the TPB or any regulator.

Access to the platform is gated on your TPB registration: we verify your number before you come on board, and the registration is monitored so a lapse is caught rather than ignored. Registration details identify your firm on the regulated reports the platform produces.

We are building a panel of independent, TPB-registered tax agents to read, edit, and endorse the strategy catalogue for each financial year, with the reviewer's name and TPB number staying with the strategy. This layer is rolling out — endorsements appear on strategy cards as reviewers come online. See the strategy reviewer panel.