Australian AI Governance in Drupal: Scoring AI Against the DTA Policy, AI Ethics Principles and Guidance for AI Adoption (Feedback Wanted)

Australian AI Governance in Drupal: Scoring AI Against the DTA Policy, AI Ethics Principles and Guidance for AI Adoption (Feedback Wanted)

If someone asked whether your AI guardrails were followed, what would you show them? That question is doing the rounds at this year's AI Government Showcase in Canberra, and it is the question the AI Empathy Evaluation module for Drupal was built to answer. Every AI response is scored, and every score is kept as a record you can show.

The module already scores responses against a catalogue of AI governance frameworks, from the EU AI Act to India's AI Governance Guidelines. Until now it had nothing for Australia. I have put together a change that adds three Australian frameworks, proposed for version 1.6. Before it ships, I would like feedback from people who work with these frameworks every day.

The three Australian frameworks

FrameworkTypeWhat it asks for
Australia's AI Ethics Principles (2019)Voluntary principlesEight principles, including fairness, privacy protection, transparency and explainability, contestability and accountability. Still the national baseline.
Guidance for AI Adoption (National AI Centre, October 2025)Voluntary guidanceSix essential practices: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, and maintain human control. It updates the 2024 Voluntary AI Safety Standard.
Policy for the responsible use of AI in government v2.0 (Digital Transformation Agency, from 15 December 2025)Mandatory for non-corporate Commonwealth entitiesAccountable officials, an accountable owner and impact assessment for each in-scope AI use case, a register of use cases, a published AI transparency statement, and staff training.

Each one becomes an editable policy in the module, with its jurisdiction, type, themes, source link, version and an as-of date, plus a short piece of guidance: a description of what a response that honours the framework looks like. For the DTA policy, for example, that is a response that is open about AI being involved, keeps a person accountable for decisions that affect people, explains the outcome in plain language, and tells people how to seek review.

How it works in Drupal

  • Choose what applies. Tick the Australian policies site-wide, or only for particular evaluation scenarios, such as a benefits or debt-notice scenario.
  • Score every response. Each AI response gets a Policy Alignment score from 0 to 100 against the policies you chose, alongside the empathy scores, with a threshold you set.
  • Keep the evidence. Every evaluation is stored as a record with its scores and the policies it was checked against. With the OpenTelemetry integration, results can also go to the monitoring tools you already run.
  • Stop responses that fall short. The empathy guardrail for the AI module can block chat responses below a threshold, before anyone sees them.

That last point matters most in government. Australia's experience with automated debt recovery showed what happens when automated decisions are hard to question. Contestability is one of the eight AI Ethics Principles, and telling people how to seek review is part of the guidance for every Australian policy in this change.

As with the rest of the catalogue, this describes alignment with each framework. It is not legal advice, and a score is not a compliance certificate.

Where I would like feedback

  1. Are these the right three? Should state frameworks be included too, such as the NSW AI Assessment Framework or Queensland's AI governance policy?
  2. Is the guidance faithful? Especially for the DTA policy: does the summary match how agencies read their obligations?
  3. Would an Australian demo scenario help? For example, an automated debt or benefits notice, scored for empathy, contestability and transparency.
  4. What evidence do agencies actually need? Could evaluation results help with a use-case register entry or an AI transparency statement, and what would that need to look like?

Comment on the issue, #3616692: Add Australian AI governance policies to the governance catalogue, or review the code in merge request !43. I will keep the issue open for a while so there is time for feedback to shape the change before 1.6.0.

Try it

composer require drupal/ai_empathy

The Australian policies are in the merge request, not in a release yet. To try them now, apply the merge request's changes to the 1.6.x branch. If you are new to the module, AI Empathy 1.4 and 1.5 covers the governance catalogue and the rest of the module, and your AI governance policy is a PDF that your prompt has never read explains why I built it.

If your agency or team is putting AI in front of the public and needs to show it is being used responsibly, I am happy to help you set this up or run a pilot evaluation. Get in touch.