Australian AI Governance Framework

Australian AI Governance Framework

Australian AI governance framework, built around the legislation a specific organisation's use of AI actually triggers.

This repository contains no source code. The service is closed-source and hosted at aiframework.com.au. This repo exists as the public metadata pointer for MCP directory listings, and as the connector's public documentation.

---

What it does

Profiles an organisation by industry, size, turnover, states of operation, AI use cases and data types, then identifies the Commonwealth and state instruments that apply to that profile specifically:

• Privacy Act 1988, including the Notifiable Data Breaches scheme
• APP 1.7 — the automated decision-making disclosure required in covered businesses' privacy policies from 10 December 2026
• Sector-specific regimes triggered by the profile, such as the Heavy Vehicle National Law, the Fair Work Act 2009 or the Security of Critical Infrastructure Act 2018

The mapping is tiered rather than a flat list. Instruments return under Applies, May apply depending on facts you hold, or Applicability turns on your own status — because the questionnaire captures a profile, not an organisation's full circumstances, and several thresholds turn on facts it does not ask for. The Privacy Act small business exemption is the clearest case: at $3 million turnover or less, whether the Act applies depends on which of the ss 6D and 6E categories the organisation falls into.

A tier with nothing in it is left out, so a given document may carry two of these headings rather than three.

Around that mapping the framework is structured on the six essential practices (AI6) in the National AI Centre's Guidance for AI Adoption, October 2025: Accountability, Impact Assessment, Risk Management, Transparency & Information Sharing, Testing & Monitoring, and Human Oversight.

The legislative mapping is specific to the profile supplied. It is not something a general answer reliably produces for an Australian organisation.

---

Endpoint

There is no authentication and nothing to provision. No API key, no OAuth flow, no account. Any MCP-compatible client can connect and call all three tools immediately.

---

Installation

This is a remote, hosted MCP server. No local installation is required.

Claude Desktop / Claude Code

{

"mcpServers": { "australian-ai-governance-framework": { "url": "https://aiframework.com.au/api/mcp.php" } } }

Any other MCP client — point it at the endpoint over Streamable HTTP and leave authentication unset.

If you have connected before and the tool names or the questionnaire's options look wrong, remove the connector and add it again. Tool lists cache client-side: the tools were renamed in August 2026, and the questionnaire's options change as the legislative register grows.

---

Tools

1. start_australian_ai_governance_framework

No parameters. Returns a session ID and a seven-question questionnaire: industry, organisation size, states and territories of operation, annual turnover band, AI use cases, data types, and an optional free-text description of the organisation.

2. submit_ai_governance_profile

Takes session_id, answers and contact.

Six answers fields are required — industry, orgSize, jurisdiction, turnover, aiUseCases, dataTypes. additionalNotes is optional. Values are validated against the option strings the questionnaire returns, so pass them through unaltered: jurisdiction takes full state names rather than abbreviations, and punctuation in option strings is significant.

contact requires name, email and organisation, and the server validates the email address. The organisation name is printed as the document's "Prepared for" heading and appears throughout the generated text, so it should be the organisation the framework is for.

Returns the session ID and a status of profile_saved. It does not return the framework.

Safe to call again on the same session: the profile is overwritten rather than duplicated, and the contact record is keyed on the email address. Re-submitting discards any framework already generated against that session, so call it again only to correct an answer.

3. get_ai_governance_framework

Takes session_id. Returns the free preview, a view_url and a purchase_url.

The document arrives in the structuredContent field framework_markdown, and in the text content block for clients that read those. It is meant to be shown to the user in full as Markdown — it is the thing they asked for, not a payload to summarise.

Generation takes around 30 seconds. That is the model producing the document, and it happens on this call. A client with a short tool timeout will give up before the endpoint answers.

---

Free preview and the full framework

The name before each bracket is the heading used in the document; the name inside is the National AI Centre's own wording for the same practice.

Instruments under Applies, and under Applicability turns on your own status, are never held back — both print in full in the free preview, Commonwealth and state alike. One tier is capped: May apply depending on facts you hold. Commonwealth and state instruments in that tier each have their own limit, and both limits are higher in the purchased framework. Where instruments are held back, the preview says so; where a profile matches fewer than the limit, nothing is capped and no such note appears.

The preview page is read-only. Printing and saving are part of the purchased framework, which also extends the link from two hours to 18 months.

There is no payment through this endpoint. The purchase_url reopens the preview with the profile already restored, alongside the purchase option, so the questionnaire is not answered a second time. Payment is by Stripe from there. Card details are never handled by aiframework.com.au.

Two complete purchased frameworks are published in full, unedited, so the paid output can be read without paying: https://aiframework.com.au/samples/

---

Do not send personal information

The profile describes an organisation, not a person. Nothing in the questionnaire asks about individuals, and the framework does not need anything about them.

This matters more over MCP than on the website. The optional additionalNotes field takes free text, and on the web form the user reads a warning before typing into it. Through the connector an agent fills that field on the user's behalf and never sees the warning. If you are building or configuring a client, carry this through:

Describe the organisation, not individuals. Do not include names, contact details, health or other sensitive information, client details, or anything commercially confidential.

Whatever goes into additionalNotes is sent to Anthropic's API for generation and stored with the session.

---

What happens to what you send

Served over HTTPS throughout. API and database credentials sit outside the web root and are never present in a browser. The endpoint applies an IP-based rate limit.

Full detail, including overseas disclosure and the Notifiable Data Breaches position, is in the privacy policy.

---

Links and retention

• Preview links expire two hours after the profile is created — both the view_url and the purchase_url. Starting again takes about a minute; the questions are the same seven.
• A purchased framework stays at its link for 18 months, then is deleted. That link is how a customer returns to the document, so it is worth saving.
• An unpaid session is deleted about two hours after it is created, when the links stop working.
• Contact records are kept for 18 months from creation, then deleted. Earlier deletion can be requested.
• A framework reflects the legislative position and the profile as at the date it was generated. It is not updated afterwards.

---

Troubleshooting

---

Scope

Output is a framework presentation only — not advice, not a recommendation, not compliance guidance. Every framework carries a disclaimer to that effect. The output is not reviewed by a human before delivery.

It does not produce AI application inventories, staff training plans, internal policies, role assignments or governance maturity ra