As of February, per a federal government report, AI had not significantly affected Australia's job market, graduate opportunities included

1

2

. This month, the Reserve Bank of Australia has flagged early signs that younger workers in AI-exposed industries are struggling to find work

1

2

. It is the first official signal that the February picture may be shifting.

What the official picture says

The July report drew a clear line between kinds of work. Jobs built on cognitive tasks, such as accountants, clerks and receptionists, were rated most exposed to AI automation. Physical work, including trades, care work and truck driving, was rated least exposed

1

2

.

Source: Phys.org

Source: Phys.org

Employer behaviour is already moving in the report's direction. Some law firms have cut graduate positions and shifted low-level tasks to AI. Some consulting firms have trimmed graduate roles in recent years too, but they say AI is not the cause. In the week before publication, Qantas said it will shift 850 jobs to India and put more money into AI

1

2

.

The evidence so far amounts to some early signs, not a confirmed trend

1

2

.

The machines are getting better at the work

The author of the articles, an academic who ran the criminal law exam studies, reports that AI's average score on university criminal law exams climbed from 52.5% in 2023 to 76.3% by 2025. On that one test set, the 2025 models on average outperformed an average university student on critical thinking

1

2

.

The author points to independent work too. In one 2024 study comparing human and AI reasoning on trick puzzles, earlier AI systems did worse than humans; newer models did better

1

2

. An OpenAI model has also recently disproved a geometry conjecture dating from 1946

1

2

. These were specific tests, the author notes, and they do not prove AI could take over a whole profession

1

[2](https://phys.org/news/2026-11-ai-job-universities.html].

What universities are being told to do

If AI can do the routine intellectual work, the remaining value of a graduate lies in what AI still cannot do unaided. In one 2026 pilot study, health workers, most of them nurses, accepted half of the deliberately wrong AI recommendations and turned down about one in six correct ones

1

2

. A UK study found members of the public sometimes leave out important details when asking chatbots for medical advice, producing answers that mix useful suggestions with misleading ones

1

2

.

The author argues the answer may be to teach both: full disciplinary knowledge and the ability to judge, direct and improve what AI produces

1

2

. The physical side of the labour market buys some time. Robotaxis, delivery robots and autonomous excavators already operate in a growing number of cities and sites. How soon robots could do a tradesperson's work is unclear, so vocational pathways currently carry lower AI-disruption risk

1

2

. Per the author's argument, an employer may have little economic reason to keep a worker who cannot improve on what the AI produces.

Source: The Conversation

Source: The Conversation

© 2026 TheOutpost.AI All rights reserved