Somewhere in Australia this year, a compliance manager will sit across from an auditor and hear a question the sector has never had to answer before. It is a question about AI assessment governance, and most registered training organisations do not yet have a confident answer to it: "Your trainers are using AI in assessment. Walk me through how assessor judgement is preserved and evidenced." Not "do you use AI." That question is already obsolete. The auditor knows AI is in the building. The question is whether its use is governed, documented, and defensible under Standard 1.4 of the Standards for RTOs 2025. Standard 1.4 has not changed what assessment must be. It must still meet the Principles of Assessment (validity, reliability, fairness, flexibility) and the Rules of Evidence. What has changed is the environment those principles now operate in. When an AI tool contributes to a marking decision, the auditor's follow-up questions become very specific:

• Who made the final judgement, the assessor or the tool?
• Can you show me where the assessor reviewed, amended, or overrode the AI's output?
• How was the assessment mapped against the rubric and unit requirements?
• Did the student see and acknowledge the outcome?
• If I pull any assessment from the last twelve months, can you produce this trail?

Most RTOs cannot answer those questions with evidence. They can answer them with intentions. "Our trainers always review the AI's suggestions" is an intention. It is not evidence. An auditor cannot audit good intentions. This is precisely the gap that ASQA has begun signalling through its own sector communications, including its guidance on using artificial intelligence compliantly under the 2025 Standards.

What Strong AI Assessment Governance Actually Looks Like

So what does a defensible answer actually look like? Effective AI assessment governance is not a single policy document sitting in a drawer. It is a working system that produces evidence as a by-product of normal marking activity. In my view it rests on four things:

1. Documented human oversight. Every AI-assisted decision carries a record showing an assessor reviewed it. Not a policy stating they should. A record showing they did.
2. A signed decision trail. The assessor's sign-off is captured per assessment, with a timestamp, so judgement is attributable to a qualified person, every time.
3. Rubric mapping. The AI's contribution is anchored to the marking rubric and unit requirements, so validity and reliability can be demonstrated rather than asserted.
4. Student acknowledgement. The student sees the outcome and confirms receipt, closing the loop on fairness and completing the evidence chain.

Notice that none of these four elements require you to abandon AI. They require you to wrap it in a defensible process. The tooling that surrounds your assessment matters just as much as the assessment itself, which is why the same discipline RTOs apply to their learning management system should now extend to how AI touches marking decisions.

Why a Ban Is Not the Answer

Here is the uncomfortable part. Many RTOs have quietly banned AI in assessment to avoid this problem. That does not solve it. Your assessors are already using these tools, often on personal accounts, with no trail at all. A ban does not remove the risk. It removes your visibility of the risk. Worse, it pushes the activity into channels your quality system cannot see, let alone evidence, when an auditor asks. The providers who will walk out of their next audit comfortably are not the ones who avoided AI. They are the ones who can put a complete evidence chain on the table for every assessment decision, AI-assisted or not. The same principle applies across an RTO's digital footprint, from assessment records to a compliant RTO website that reflects the same standard of governance. That is the standard the sector is moving toward, whether we are ready or not. AI assessment governance will soon be treated as a baseline expectation rather than a competitive advantage, and the RTOs that build the evidence habit now will be the ones spared a scramble later. How is your RTO documenting this today? Genuinely curious what approaches are out there, because right now everyone seems to be solving this in isolation. Get in touch with me to discuss how your institution can adopt to AI assessment governance.