AI, in your work.

Access to tools and demonstrated competence

The announcement OpenAI published on November 5 brings together collaborations with Australian universities to incorporate AI into education. Two cases, UNSW Sydney and the University of Sydney, help examine a specific question: how to combine practice with these tools with assessments that establish what each student can do.

University sources describe both technology rollouts and assessment criteria. For training designers, that distinction provides a starting point: define the capability to be assessed before deciding what assistance is allowed.

What access is available and what remains planned

On September 11, UNSW explained that it was extending ChatGPT Edu to Higher Degree Research candidates and casual academics, building on access already available to academic and professional staff. That update placed access for all students at the start of the 2027 academic year.

Sydney's FAQ schedules general student access for Semester 1, 2027. It still describes the student pilot as planned for October 2026, without confirming completion or providing results. The rollout prioritises staff.

The university's September announcement pairs access with training and support. It also retains Cogniti as the preferred classroom tool, with ChatGPT Edu supplementing it.

Advanced features, including agents and connectors, require individual assessments before being enabled; dates and functionality may change, according to Sydney's FAQ.

UNSW: allow, prohibit or require AI according to the learning goal

On October 13, UNSW published an assessment framework with three modes. Program teams distribute these modes across the curriculum to develop and assess different capabilities.

Sydney: secure and open assessments

Sydney describes a two-lane assessment approach: supervised, in-person assessments to establish learning outcomes, and open assessments. The combination is planned across programs; individual units specify their tasks and instructions.

Its responsible-use guidance allows AI in open assessments and requires students to acknowledge its use. In secure assessments, AI is generally prohibited unless expressly authorised in the unit outline. Access to a tool does not change those conditions.

Nexo's interpretation: questions for training managers

Our interpretation is that these cases offer useful questions for designing workplace training. Applying them at work requires adapting objectives, tasks and criteria; this is neither advice from the universities to businesses nor evidence validating that transfer.

• What knowledge should participants demonstrate independently, and in what situation will they need to apply it?
• Which task requires learning to collaborate with AI, and how will that capability be observed?
• What evidence of the process will be required: verified sources, identified errors, justified changes or explanations of decisions?
• How will participants subsequently demonstrate that they can solve a variation of the problem?

Synthetic example: resolving a project issue

Nexo proposes the following fictional exercise, with no client data or real results. The starting material would be a simulated case file containing a delay, task dependencies and team messages. The objective would be to assess diagnosis and judgement when preparing a response.

• First, participants independently identify the likely cause of the delay, supporting evidence and missing information, without AI.
• Next, they use AI to prepare a response to the team and check each proposal against the case file. They submit a revised version and explain what they accepted, corrected or rejected.
• Finally, they answer questions about their decisions and solve a small variation of the case without assistance.

What would need to be measured

For this exercise, we propose separately assessing diagnostic accuracy, evidence verification, justification of decisions and subsequent performance without assistance. The quality of the final text would be only one part of the assessment.

The sources consulted describe policies and implementation plans, but do not provide a controlled comparison establishing learning or workplace productivity gains from these two rollouts. Announcements of access and the existence of assessment rules should be read within that limit.