Last week, we shared a Stanford study that found AI hiring tools rejecting Black and Asian applicants at disproportionate rates across 156 companies that all happened to be using the same vendor’s screening algorithm. One vendor. One blind spot. A structural wall across an entire labour market.
Australia has not run that exact study yet. We do not need to. Three separate pieces of research already point at the same failure from different angles. And when you lay them next to each other, the picture is not just uncomfortable. It is a crisis that AI is now turbocharging.
Start with the headline claim.
85% of companies say they have adopted skills-based hiring. Degree requirements removed. Job posting language updated. Public commitment made. Sounds like progress.
Except when the Harvard Kennedy School and Burning Glass Institute checked what actually happened to who got hired, only 1 in 700 hires were impacted by the degree requirement removal. Less than two tenths of one percent. The posting changed. The decision did not.
Now bring it home.
The Australian Information Industry Association found that only 1% of Australian tech graduates are assessed as immediately work ready. Not 1% are unqualified. 1% are assessed as ready, using whatever method the assessor happens to be using. Meanwhile Jobs and Skills Australia projects that more than 90% of employment growth through 2035 will sit in occupations requiring post-secondary qualifications. And the Graduate Outcomes Survey shows full-time graduate employment falling even as the broader economy keeps adding jobs.
| Bachelor degree or higher | 54% |
|---|---|
| Diploma or advanced diploma | 11% |
| Certificate III or IV | 9% |
| Certificate II or III | 20% |
| Certificate I or secondary school | 6% |
Source: Jobs and Skills Australia, Employment Projections (December 2025), Table 3. Redrawn by Gradstack, CC BY 4.0 © Commonwealth of Australia.
The posting changed. The outcome did not. The economy needs more of exactly the people getting filtered out at the front door.
Then AI entered the picture.
Here is where the original broken system gets a dangerous upgrade. When humans ran a credential-biased, resume-led, high-noise-low-signal screening process, they were at least limited by bandwidth. A recruiter can only reject so many people in a day.
AI has no such limit.
The same structural bias that produced 0.14% skills-based hiring outcomes is now encoded into screening tools running at scale, across thousands of roles simultaneously. And it is not just the screening tools. AI-generated resumes are flooding the top of the funnel, optimised to beat keyword filters, trained on the same historical hiring data that already excluded the same people. So the system is now processing a higher volume of noise, faster, with greater confidence that it is being objective, because an algorithm said so.
The Stanford finding is not a US problem. Australian employers use the same US-built AI hiring tools in their day-to-day workflows. The same vendor risk. The same blind spot. Running daily.
This is not a national talent shortage. It is a measurement failure at speed.
Australia does not have a capability gap. It has a signal gap. The tools doing the screening, whether that is a resume keyword filter, a degree requirement quietly still in the background, or an AI ranking system trained on who got hired in the past, are not measuring whether someone can do the job. They are measuring whether someone resembles who already has the job or fits the brief, and whether their resume was optimised well enough to clear a filter that should not exist.
Those are not the same thing. The 0.14% figure proves it. The Stanford study proves it. The 1% work-ready assessment figure proves it.
The fix is not asking employers to care more.
Most already say they want this fixed. That is exactly how you get 85% adoption sitting next to a 0.14% outcome. Caring is not the constraint.
The fix is verified capability signals that cannot be gamed, AI-generated, or excluded by a biased filter. Signals that travel with the person, not with the institution they came from. Signals that make it structurally harder to fall back on credential shortcuts because the actual evidence of what someone can do is already in front of you.
That is what the VCI® (Verified Capability Index) and VEI® (Verified Employability Index) exist to do. To fix the measurement layer. Because if the signal is verified, portable, and capability-based, the system has to respond to it. The credential bias loses its grip. The AI filter has something real to work with instead of noise.
The posting changed 85% of the time. The hiring outcome changed 0.14% of the time. The difference between those two numbers is the gap Gradstack is built to close.
We will not stop until every capable person gets the signal they deserve. 💚