Australia's First Gen AI Capacity Study: 79% of Workers Face Low Risk

If you work in Australia and you've spent the last year quietly worrying that AI was coming for your job, here's a number worth knowing: 79% of Australian workers face low or very low risk of automation from generative AI, according to the country's first national Gen AI Capacity Study. That's not a survey of opinions — it's a structural analysis of 358 occupations done by the federal Jobs and Skills Australia (JSA) agency. And it tells a story that's almost the opposite of the headlines.

The full report, Our Gen AI Transition: Implications for Work and Skills, was released in August 2025 with occupation-level data following in September. It's the largest government-led labour-market study of generative AI we've seen from any country so far — and the conclusions are deliberately, almost uncomfortably, measured.

Here's what they actually found, and why it matters for whoever's reading this from a desk in Sydney, Melbourne, Perth, or anywhere else.

The headline number, in plain English

JSA scored every task inside every occupation in the Australian standard (ANZSCO) on two separate axes: augmentability (could Gen AI help the worker do this task better or faster?) and automatability (could Gen AI do the task without the worker?). This two-axis approach matters, because most coverage collapses them into one — and that's where the panic comes from.

When you keep them separate, the picture changes. JSA's analysis found that augmentation generally outweighs automation [Fact] across the Australian workforce. In their words: Gen AI is "more likely to enhance workers' efforts in completing tasks, rather than replace them."

The 79% low-risk figure refers to workers whose roles are dominated by tasks that either Gen AI can't do well, or where Gen AI's contribution sits clearly in the augment-not-replace bucket. The remaining 21% isn't a death sentence either — it's workers with significant exposure to automatable tasks, who will see meaningful workflow restructuring rather than wholesale displacement.

For context, the OECD's earlier global estimates put roughly 27% of jobs in advanced economies at "high risk" of automation from AI generally. JSA's narrower 21% is consistent with that ballpark once you carve out the specifically generative-AI slice and adjust for Australia's industry mix (mining, agriculture, and services-heavy, with a smaller manufacturing share than the OECD average).

Where the real exposure is

The vulnerable layer in JSA's analysis isn't where most people guess. It's not truck drivers, not tradespeople, not nurses. The concentration is in routine clerical work — data entry, basic record-keeping, simple correspondence, repetitive document handling. ANZSCO Skill Level 5 clerical and administrative support roles take the biggest hit.

But here's the twist JSA flags repeatedly: even in those exposed occupations, full replacement is rare. What's far more common is task-level substitution inside a job that otherwise remains. A claims processor whose document review is now 40% AI-assisted is still a claims processor; the role contracts in some directions and expands in others (exception handling, complex case judgment, AI output verification).

Professionals and managers at ANZSCO Skill Levels 1 and 2 — accountants, lawyers, engineers, project managers, mid-senior public servants — show the highest augmentation scores in the entire study. The interpretation matters: these are people whose jobs contain a lot of repetitive cognitive admin (drafting, summarising, formatting, basic analysis) that Gen AI handles well. Freeing that time up doesn't shrink the role; it expands the share of the job spent on judgment-heavy work clients and employers actually pay senior salaries for.

In other words, the same technology that makes a routine clerical job more precarious tends to make a senior knowledge-worker job more valuable. That's a distributional issue JSA itself raises explicitly — and it's one of the most under-discussed implications of generative AI anywhere in the global labour-market literature.

Why JSA's number is more credible than the doom forecasts

Three things separate this study from the wave of "X% of jobs at risk" reports that have flooded the discourse since 2023.

First, it's task-level, not occupation-level. Most early forecasts (Frey & Osborne 2013, the original viral source of "47% of jobs at risk") scored whole occupations as automatable or not. That's a coarse instrument. JSA breaks every occupation into its component tasks (from ANZSCO and Australian job-ad data) and scores those individually. A "marketing manager" isn't replaced by AI; the 30% of their week spent on copy drafting and analytics reporting is partially automated, while the 70% spent on strategy, stakeholder management, and budget negotiation isn't.

Second, it separates augmentation from automation. Earlier studies — including some published as recently as 2024 — quietly conflate the two. If a task can be done with AI assistance, those models often counted it as "exposed" and rolled it into the headline risk number. JSA's two-axis framework refuses to do this, which is why their numbers come in lower than a lot of the McKinsey/PwC-style consultancy estimates [Claim].

Third, it's grounded in actual Australian labour-market data, not extrapolated from US occupation structures. This matters more than people realise. The US has a much higher share of office-administrative employment than Australia (about 18% vs. roughly 13%) and a smaller share of trades and technical occupations. Importing US-based exposure scores onto an Australian workforce overstates risk by mechanically inflating the categories most at risk [Estimate].

What this means if you're sitting in one of those exposed roles

Let's be concrete. If you're in administrative support, data entry, basic accounting clerking, call-centre routine handling, or junior paralegal document review — the JSA data says your role is changing, not vanishing. The honest version of the advice goes like this:

The repetitive parts of your job will shrink. The judgment, exception-handling, communication, and AI-supervision parts will grow. Whether your specific role survives depends on whether you move toward the second category fast enough, and whether your employer chooses to redeploy or shed the headcount the productivity gain creates.

That second condition is where the JSA report gets unusually candid for a government document. It notes that labour market dynamism — the ability of workers to move between roles, sectors, and skill levels — is one of the three pillars of a healthy Gen AI transition (alongside exposure analysis and adaptation). Australia's overall job mobility has been declining for a decade, which is the structural risk JSA quietly flags as the actual problem to solve. The AI isn't the threat. The lack of pathways out of shrinking roles is.

This is the kind of nuance you almost never see in tech-vendor or consultancy reports, which tend to push either "everything's fine, lean into the tools" or "reskill or die." JSA's view sits in the harder middle: the technology is real, the exposure is real, the displacement risk is real for a specific sub-population, and the policy and personal response that actually works isn't "learn to prompt" — it's structural mobility, sector-specific transition support, and active labour-market policy.

What's missing from the report (worth knowing)

A few honest caveats. JSA's analysis is occupation-level structural exposure, not real-time employment data. It's a map of where the risk concentrations are, not a forecast of how many Australians will lose their job in 2026 or 2027. The agency is explicit that the actual employment impact depends on adoption pace, firm-level decisions, and policy response — none of which the exposure data captures.

The study also leans heavily on tasks that are visible in formal job descriptions and ANZSCO definitions. Tacit knowledge, informal coordination work, and the kind of judgment that doesn't show up in a job ad get systematically under-weighted in any analysis built on this kind of data. That probably means the augmentation-not-automation finding is, if anything, conservative — actual jobs contain more non-automatable work than their formal task lists suggest.

And finally: this is a snapshot of Gen AI as it existed in early-to-mid 2025. Capability is moving fast. JSA notes the study should be re-run as the technology matures, and the agency has committed to ongoing monitoring rather than treating the August 2025 report as the final word.

The takeaway worth carrying around

The number that should stick: 79% low risk, 21% meaningful exposure, and within that 21%, the realistic outcome for most workers is task restructuring inside a continuing role, not unemployment. The people who should be most worried aren't the people in the most "AI-exposed" occupations — they're the people in exposed occupations who also work for employers with weak track records on retraining and redeployment, in regions and sectors with limited job mobility.

That's a much smaller, more specific population than "everyone whose job has any AI-doable tasks." It's also the population that public policy can actually do something about — which is, ultimately, why a government agency produced this kind of analysis in the first place.

For the rest of us, the JSA report is a useful reset against the doom-by-headline coverage. The technology is real. The displacement risk is real for a specific slice. But the default outcome for the majority of workers is augmentation, not replacement — and the studies that keep finding otherwise are mostly built on methodologies that JSA's two-axis framework was designed to correct.

---

Sources

• Our Gen AI Transition: Implications for Work and Skills (Overarching Report) — Jobs and Skills Australia, August 2025
• Occupation Data on AI Exposure — JSA, September 2025
• Industry Data on AI Exposure — JSA, September 2025

_AI-assisted analysis. Primary source: Jobs and Skills Australia (Australian Government)._