The AI decision organisations can’t ‘make by default’

The debate on AI governance has grown beyond typical technology regulation parameters to an all-encompassing business decision, Robert Walters has found.

Further to ongoing coverage over the deficiencies in governance regarding professional AI adoption, a letter penned by the Australian Prudential Regulation Authority (APRA) on 30 April 2026 has reframed the debate between AI and human capacity.

According to recruitment and business solutions company Robert Walters, APRA’s wording and language is more closely associated with workforce management over IT policy, as is its insistence that technology be owned and moderated at all times by humans.

These specifications are “not a technology specification, it’s a job description”, said Michelle Christie, senior commercial director at Robert Walters Australia.

And while it is generally accepted that humans’ ability to form complex judgements ranks higher than that of AI, Christie suggested that this doesn’t mean humans always perform this correctly.

She said: “The harder question is whether the people function has the judgement, literacy, and cross-functional authority to actually govern AI day to day, not just sign off on it.”

As revealed in the Robert Walters e-guide “AI governance: the new accountability agenda”, institutions are generally fractured on AI governance due to disagreements on its very nature.

Those who understand it as a technology problem will prioritise technical depth and model monitoring, thereby deferring to the CTO or CDO, while a CRO or CPO is usually consulted when organisations tackle it from a human angle, prioritising accountability, oversight and human judgement regarding AI decisions.

However, Robert Walters expressed that neither solution tackles the issue fully, accounting for a rise in chief AI officer roles as organisations struggle to address both sides of the issue.

George Clarke, head of growth at Robert Walters Australia, said: “The institutions getting this right aren’t choosing between a human-led or technology-led model, they’re working out how to put real technical capability inside an environment with a robust set of controls governing it.”

This approach, Robert Walters purports, addresses the technical reality that you cannot govern what you can’t monitor. This is especially true if AI is making consequential financial decisions – and in such cases, a human within the governance chain must understand model drift, training data and failure modes to the degree that decisions can be made regarding necessary escalation.

As the organisation surmised, accountability only exists in practice if the human owner has the required technical fluency.

Clarke said: “Whoever ends up owning AI governance inherits a regulatory relationship, not just an org chart line.”

Robert Walters indicated that most institutions will grapple with this tension over the next year, with some likely to experiment with roles that combine capabilities of the CTO and CPO, while others may supplement existing frameworks with more relevant technical or governance capability.

Whether a general consensus emerges is yet to be seen – but what is clear is that failing to make a decision will be as consequential as making the wrong choice. Either way, it is apparent that the earlier these considerations are made, the better.

Clarke said: “Getting the structure wrong now is expensive to unwind later.”

Want to see more stories from trusted news sources?
Make HR Leader a preferred news source on Google.