Tech in Gov 2026: Judgement in AI — building frameworks for responsible AI

Summary:  At Tech in Gov 2026, Britt Spyrou, Director of the Enterprise Project Management Office at the Department of Parliamentary Services, ran an interactive workshop on building frameworks for responsible AI. She introduced the AI governance stack and a four-step assessment process including the eight AI ethics principles. She then had the audience apply them to two real-world use cases.

Date:
27 August 2026
Author:
Phillipa Martin
 
Phillipa Martin is Salsa’s Rules as Code Practice Lead and Content Strategist.

Workshop: Judgement in AI — building frameworks for responsible AI

Workshop — 5 August 2026, 1 pm. Britt Spyrou, Director, Enterprise Project Management Office, Department of Parliamentary Services.

This Day 2 workshop took a hands-on approach to responsible AI, asking the audience to assess real use cases rather than just talk about principles.

An interactive start

Spyrou opened by polling the room on where people worked — around half were from federal government agencies, with the rest from state government, universities and the private sector.

Using an interactive slide system called AhaSlides, she then asked the audience to describe in two words how they were feeling about AI. The most common word:

"Excited"

She then asked how they felt about the APS using AI, the more common responses were along the lines of:

"Impatient"

"About time"

Proportionate fit

Next, Spyrou set the policy context, referencing the Policy for the responsible use of AI in governmentExternal Link .

Her point was about proportionate fit: standards are emerging, but operationally you still need governance matched to real-world impact. That framed the two questions the workshop kept returning to — what is the real-world impact, and what does proportionate governance actually mean?

Three pillars of responsible AI

She described responsible AI as resting on three pillars:

  1. Enable opportunity
  2. Build public trust — earning it, rather than assuming it
  3. Adapting to change

The governance stack

Spyrou then introduced the governance stack, which sets out where responsibility sits at each level:

  • “National setting (Office of AI)External Link — national standards, data-centre and AI training direction
  • APS/agency setting — accountable officials, transparency, strategy, registers, impact assessment and training
  • Use case setting — AI purpose, affected people, influenced decisions and the evidence needed
  • Individual setting — check the task, verify outputs, protect information, keep records, and escalate as impact increases

A four-step process

To assess the use of AI, Spyrou detailed a four-step process:

  1. Practical impact — what effect does this actually have in the real world, including who is affected?
  2. Ethical design — assess against the eight AI ethics principles.
  3. Existing guardrails — are existing controls already working to reduce the risk? These include data governance, privacy controls, records management, delegation and financial controls, and audit and assurance.
  4. Proportionate governance — what's the gap? Match AI-specific oversight to the impact:
    • Low — existing controls, light touch
    • Medium — deliberate oversight, shared visibility
    • High — formal assessment, clear escalation

The eight AI ethics principles

Next, were the eight AI principles:

  1. Wellbeing
  2. Human-centred
  3. Fairness
  4. Privacy
  5. Reliability
  6. Transparency
  7. Contestability
  8. Accountability

To see the process in action, Spyrou took us all through two use cases, asking us to consider the four-step process.

Use case one: AI-generated briefing summaries

The first use case was public sector employees using AI to create summaries for a briefing document for the SES.

Using interactive polling slides, Spyrou asked the audience to scan a QR code and vote from their phones on whether the impact was low, medium, high — or grey. She then asked which of the AI ethics principles were most relevant to the use case, and worked through the results, taking questions and comments from the room.

Next she asked which existing controls already apply here, with options including:

  • Data quality controls
  • Privacy safeguards
  • Security and access controls
  • Record-keeping requirements
  • Delegations and financial controls
  • Assurance and audit

Finally, the audience voted on the proportionate governance controls needed, choosing from:

  • Basic controls (privacy, security and similar)
  • Record the use case
  • Impact assessment
  • Assign owners
  • Set oversight
  • Training
  • Monitoring
  • Review
  • Incident management procedures

These were presented as a sliding scale, with options further to the right generally applying to higher-risk situations.

Use case two: AI triage

The second use case was an AI triage process, with an AI system used at the front end of an eligibility process for grants, programs or services, before a human assessor reviews applications. The AI was doing three things:

  • Screening applications
  • Flagging higher-risk or lower-priority cases
  • Prioritising the order for review

Spyrou took the audience through the same assessment process and polls — practical impact, most relevant ethics principles, existing controls that already apply, and the proportionate governance controls needed.

Three key takeaways

Spyrou closed with three takeaways:

  • Opportunity — there's a huge opportunity here. Use AI to improve services while keeping its use lawful, ethical and accountable.
  • System and self — it's a combination of both. The system sets the guardrails, and people review, approve and escalate.
  • Proportionate — use existing controls first and lift oversight as risk, scale or impact increases.

Her final point: responsible AI is about keeping public trust while we use it well.