ACS Canberra hosted the AI Masterclass for Government Leaders and developed the Decision Edge Framework. Executive committee member Eric Nguyen reflects on how leaders can prevent AI logic drift and make AI-assisted decisions defensible.

AI can quietly shift the reasoning behind your decisions before anyone notices – until it suddenly becomes someone else's problem.

We call it AI logic drift: the quiet erosion of causal reasoning, context, and intent as a decision moves through the decision-making chain and various stakeholders.

In the process, the original intent – what the decision was actually trying to achieve – gets lost somewhere.

Then your decision poses a critical question: "Is it logical, or compliant only?"

Confidence is not defensibility

Assistant Minister Patrick Gorman made the accountability standard plain earlier this year: you are personally accountable for AI errors. The outputs are yours.

The evidence supports the urgency.

Knowledge workers using AI are 19% less likely to find the correct answer when tackling complex tasks that require nuanced judgment.

A landmark Harvard and BCG study of 758 professionals uncovered a stark reality – unless the human in the decision chain has a structured framework to actively catch "logic drift," delegating deep thinking to AI leads straight to defective decisions.

The 2026–27 budget embeds AI deeply into housing approvals, medicine assessments, and the National Construction Code.

Every agency had a deadline of this month to appoint a Chief Artificial Intelligence Officer (CAIO), drawn from existing staff, alongside everything else those executives already carry.

The mandate creates the role; the next step is building the professional capability to match it.

As ACS President Beau Tydd notes, managing the risks around "governance, accountability, workforce readiness, and public trust" is essential as AI capability scales – across government and industry alike.

Confidence in a decision is not the same as defensibility of a decision. That distinction matters as much in a corporate boardroom as it does in a ministerial brief.

Right now, for most leaders, there is no structured way to close the gap between the two.

Better than copy and paste prompts – a practical, reusable approach

The defensibility challenge isn't hypothetical.

It is exactly what 30 senior leaders – including public sector SES Band 1 and 2 officers, inaugural CAIOs, and senior corporate executives – discovered firsthand at the ACS AI Masterclass for Government Leaders.

"Roll out of AI is not a system roll-out," one participant put it plainly. "We need to upskill people for their work."

Most conversations about AI start from the same place: AI is the capability, and leaders just need to catch up through basic literacy and training.

We started from the opposite assumption – your wisdom, your institutional knowledge, and your professional judgement – that is the asset.

The Decision Edge Framework uses AI to reveal whether those assets are actually driving your decisions or getting diluted along the way.

The prompts attached to the framework test three things:

  • Whether the logic underneath a decision is genuinely causal or merely correlational
  • Whether the context has drifted from the original intent
  • Whether that intent survived the decision chain at all.

What came back from the exercise was not a verdict. It was a revelation.

What the prompts revealed — and what the room didn't expect

As the designer of the framework, I watched the room undergo a deliberate shift in perspective.

Thirty highly capable leaders needed to alter their mental model, moving from viewing AI as a ‘calculator’ to viewing it as a ‘logic auditor’.

Their role is not to verify every supporting detail in the chain. It is to audit whether the logic holds.

Once that conceptual scaffolding was in place, something shifted.

The prompt sequence made visible what had previously been implicit. Not just whether the logic was causal, but whether the context had moved as the proposal travelled through the organisation and whether the original intent was still intact by the time it reached the room.

"Never thought of working with AI in the team and organisation context," one participant said. "The orchestration of all, not individual brilliance."

That's the shift this piece is really about. It is not about AI as an isolated personal productivity tool.

It is about AI as an orchestration mechanism that reveals and multiplies human capability when it is structured to do so.

The asset is already in the building

Your institutional knowledge, your experience, and your professional judgement are real. They are also invisible: assumed rather than tested, implicit rather than structured.

Decision Edge doesn't create that capability; it makes it legible.

As I wrote to participants afterwards: "The prompts are only as valuable as the quality of the conversation you bring to AI through your own experience, judgement, and insights."

Run the sequence alone before a significant call. Run it with your team. AI reveals whether the logic holds, whether the context drifted, and whether the intent survived.

If it holds, you sign off with confidence. If it doesn't, you caught it before anyone else had to.

At the end of four hours, after the roundtables and the case studies, one participant offered a single word to sum up the experience: Trust.

Not a framework. Not a certification. Not a budget line.

The work has started, and the room proved it can be done.

Panel on accountability — what does 'good' look like?

Dr Paul Hubbard, Assistant Secretary of the AI Delivery and Enablement branch at the Department of Finance, summed up his answer in nine words: "AI changes how we work, not who is accountable."

He's right. And we're relearning what that means – how we delegate, and how we value quality over speed.

If AI gets us to 50% of a task in a tenth of the time, the remaining effort must shift heavily into validation, assurance, and delivering the final outcome at full quality.

Anthea Roberts, founder and CEO of the award-winning Dragonfly Thinking, came at it from outside the framework. Her view from the panel on accountability was direct:

"I use AI in all my work, but accountability stays with me for the outcomes of my team and my AI agents.

“It is AI augmentation, which is fundamentally different from task automation. It raises the bar for judgement, learning, and assurance."

Raises the bar. Not lowers the barrier. That difference matters.

The road ahead — professionalisation of AI leadership

This is not a one-off training event.

Professionalising AI leadership means building something that compounds – a community of senior leaders across government and industry who share a common language for accountability, return to test their reasoning together, and bring their peers with them.

The classroom becomes a network. The network becomes a standard.

That's what ACS is building, drawing on its professional standards and workforce development expertise to support leaders navigating a landscape that won't stop moving.

It is not a credential to collect, but a professional community where making AI-influenced decisions defensible becomes a shared, daily practice.

Eric Nguyen is an ACS Canberra branch executive committee member, IEEE Responsible AI Lead Assessor, and Gartner Peer Ambassador. He developed the Decision Edge Framework and facilitated the AI Masterclass for Government Leaders series.