01
Do students opt in, and can they stop at any time?
Why it matters. Unsolicited messages erode trust, and the Spam Act 2003 applies to commercial electronic messages.
What good looks like. No message before opt-in, consent recorded per student, and STOP honoured immediately.
02
Are students told they're talking to AI?
Why it matters. Transparency is one of Australia's AI Ethics Principles, and students deserve to know who they're confiding in.
What good looks like. The AI is disclosed at the start, never claims to be human, and staff who step in say so.
03
Who approves what a student receives?
Why it matters. A model can be fluent and wrong. Accountability for what an institution says to a student should sit with a person.
What good looks like. Every AI-written message approved by a named staff member, with the approval recorded. Any exceptions are fixed, pre-approved content.
04
What happens when a student discloses they're unsafe?
Why it matters. This is the moment that matters most, and the one where AI should do the least.
What good looks like. Safety signals go straight to a person, AI drafting stops, and any automatic reply is fixed wording your wellbeing team has approved.
05
Does the AI make decisions about students?
Why it matters. Automated decisions about people need a high bar of fairness and contestability.
What good looks like. Risk flags are suggestions with stated reasons. Staff decide what happens next.
06
What is it allowed to talk about?
Why it matters. Outreach framed around what students owe (fees, attendance, enrolment) changes the relationship from support to compliance.
What good looks like. A defined, closed scope of student needs, with institutional matters routed to the right team rather than recorded.
07
What data does it use, and what doesn't it?
Why it matters. The Australian Privacy Principles require collecting only what's reasonably necessary.
What good looks like. A minimal student list, no access to grades or financial records, and outbound data limited to a published set of fields.
08
Where is data stored and processed, including by the AI model?
Why it matters. Many AI services route requests offshore by default, and public institutions often have residency obligations.
What good looks like. Every sub-processor listed with its location, and model inference pinned to Australia by policy, not just configuration.
09
Who at the vendor can see student data?
Why it matters. Support access is a common gap in vendor assessments.
What good looks like. No standing access, time-limited sessions with a stated reason, and an access log the institution can see.
10
How will you know it's working?
Why it matters. Engagement isn't the same as impact, and before-and-after comparisons can mislead.
What good looks like. A comparison group agreed up front, results with confidence intervals, and outcomes measured by the institution.
Frameworks worth reading
- Australia's AI Ethics Principles (Department of Industry, Science and Resources)
- Voluntary AI Safety Standard (Department of Industry, Science and Resources, 2024)
- OAIC guidance on privacy and the use of commercially available AI products (2024)
- TEQSA resources on generative AI in higher education
- Privacy Act 1988 (Cth) and, for Queensland public institutions, the Information Privacy Act 2009 (Qld)
See how Retento answers each of these in our Trust Centre.