More than a third of Australian organisations now allow AI to execute decisions or take actions without human approval, according to Deloitte's 2026 State of AI in the Enterprise report — a survey of 3,500 senior leaders across 25 countries. The returns are real: revenue gains have climbed from 20% to 27% of organisations in under a year. But 18% of Australian respondents say accountability is unclear when an AI system produces a negative outcome. AI is gaining authority faster than organisations are building the structures to govern it.
This isn't an abstract governance debate. It's a practical problem that's landed on the desks of Australian business leaders right now. As AI shifts from recommending actions to executing them — processing invoices, triaging customer requests, adjusting pricing, flagging compliance issues — the question of who owns the outcome when it goes wrong is no longer theoretical. Deloitte's findings suggest many organisations haven't answered it yet.
What the survey actually found
The headline statistic deserves careful reading. Deloitte's Australian data shows 38% of organisations allow AI to execute decisions without human approval to a moderate degree — but none report doing so to a significant degree. Australian organisations are granting AI autonomy in measured steps, not handing over the keys entirely.
The picture elsewhere in the business is more encouraging. IT and cybersecurity leads adoption, with 62% of Australian organisations running AI at scale in that function, ahead of the 54% global average. Marketing, sales, and customer service sits at 34% at scale, roughly matching the global 38%. Strategy and operations remains more experimental: 69% are piloting, but only 23% have reached scale — well behind the 38% global figure.
"What we're seeing now is a clearer picture of what drives that value," said Stu Scotis, Deloitte Global Agentic AI Leader. "Organisations are measuring returns, becoming more disciplined about AI costs and strengthening the foundations around the technology."
Work itself is being restructured. Thirty-nine percent of Australian respondents are redesigning several key processes around AI, and 73% expect AI to handle the majority of tasks in routine knowledge work within three years. But the survey found about half report only a minor change in decision-making authority — suggesting the organisational hierarchy around AI hasn't shifted as fast as the technology.
The governance gap is the real story
The most consequential finding isn't how many organisations are using AI. It's the disconnect between AI's growing authority and the oversight structures around it.
Globally, 40% of companies allow AI to execute decisions without human approval to at least a moderate degree — slightly above Australia's 38%. But the governance infrastructure is patchy. Among companies running multiple AI agents in the same workflow, half either have no formal governance or rely only on basic oversight. Just 9% say those workflows are fully governed.
When things go wrong, accountability is fragmented. Globally, 36% of companies hold the owning team responsible, 24% assign it to a central governance function, and 10% place it with the employee who prompted the system. For 23%, accountability is simply unclear.
"The issue is no longer just whether governance exists but whether governance matches the authority companies are giving AI," said Dr Elea Wurth, Deloitte Lead Partner for Trustworthy AI in Asia Pacific. "Governance must become part of the company's operating model rather than a review process added at the end."
A separate ABBYY study of 1,200 senior managers corroborates the accountability confusion. Australian leaders are split almost evenly on who should bear responsibility when AI goes wrong: 27% say the organisation using it, 27% say it should be shared with the vendor, 17% point to the vendor alone, 15% to the end user, and 14% to regulators. There is no consensus.
Australia is doing better than most — but not well enough
There's a nuance in the data that's easy to miss. Australia's 18% accountability-unclear rate is actually lower than the 23% global figure. And Australia leads the world in human-in-the-loop practices: 61% of Australian leaders say humans review all important AI decisions before action is taken, the highest of any market in the ABBYY study.
But that still means 26% of Australian organisations using AI lack what the ABBYY research describes as meaningful human oversight — 23% review decisions only after implementation, and 13% have no formal approach at all. For companies deploying AI in live customer-facing or financial processes, that's a significant exposure.
This tracks with what we've seen in APRA's recent governance letters to banks and insurers and the broader pattern of Australian businesses discovering AI audit gaps. The regulatory pressure is building, and the organisations that haven't clarified accountability now will be the ones scrambling when mandatory frameworks arrive.
What this means for your business
If you're running a 10-to-50-person company, the Deloitte survey might feel like it's about enterprises operating at a different scale. It's not. The governance gap applies at every level — arguably more so at smaller ones, where a single AI system might touch more of the business and fewer people are watching it.
"More than a third of Australian organisations already allow AI to execute decisions or actions without human approval to at least a moderate degree, but one in five say accountability is unclear when AI produces a negative outcome," said David Alonso, Deloitte National AI Market Lead. "As AI takes on more responsibility, being clear about who can intervene and who ultimately owns the outcome will be critical to maintaining that confidence."
The practical takeaway is straightforward. For every AI system that takes action in your business — whether that's sending emails, processing applications, adjusting schedules, or triaging support tickets — you need clear answers to three questions: What can it do without asking? Who reviews it when it gets it wrong? And who's accountable to the customer or regulator when that happens?
Most organisations, according to this research, haven't answered all three.
What to watch
The survey found 86% of Australian organisations expect their AI capability to increase moderately or significantly over the next 12 months, with 46% anticipating investment increases of 6-19%. That means the gap between AI authority and governance is likely to widen before it narrows — unless organisations build accountability into their AI operating model now.
Watch for Australia's mandatory AI governance frameworks, which are already taking shape at the regulatory level. The organisations that proactively define decision rights, escalation paths, and ownership for AI outcomes will be ahead of the curve. The ones that don't will find out who's accountable the hard way — when something goes wrong and nobody can answer the question.
Sources
- Australian organisations report growing AI returns — Deloitte Australia
- State of AI in the Enterprise 2026: Built Differently (PDF) — Deloitte AI Institute
- State of AI in the Enterprise 2026 (US report) — Deloitte US
- AI moves beyond pilot phase for Aussie businesses — Financial Standard
- ABBYY Study: Australian Leaders Disagree Who's Responsible for AI — ABBYY
- Australian business leaders split on AI accountability — IT Brief Australia
