ASQA finalises responsible AI principles without new requirements

ASQA published its final 5 Principles for the Responsible Use of AI in VET on 30 July 2026.

Updated 31 August 2026 to include ASQA's assessment case study and observations about AI-drafted regulatory submissions.

The important regulatory point is what ASQA says the principles do not do: they introduce no new regulatory requirements. This is reviewed guidance about applying existing obligations when an RTO uses AI, not an amendment to the 2025 Outcome Standards.

What the principles cover

The five principles focus on:

• governance and accountability for AI use;
• human oversight of decisions and outputs;
• privacy, security and reliable records;
• fairness, accessibility and learner protection; and
• alignment with training-product and assessment requirements.

ASQA associates the guidance with Outcome Standards 1.1 to 1.4, 2.2 to 2.6, 3.2 to 3.3 and 4.1 to 4.3, as well as Compliance Requirements clauses 9 and 20.

Those associations explain where existing obligations matter when AI is used. They do not rewrite the controls or make every AI tool appropriate for every training or assessment activity.

ASQA's assessment case study

ASQA IQ for August 2026 describes an RTO testing an AI avatar for a workplace-conversation assessment. The review found that students could trigger favourable responses, speech recognition disadvantaged some accents and the avatar did not reliably capture non-verbal evidence. The RTO kept the avatar for training and practice, but used assessor observation, professional discussion and a role play with people for the assessment evidence.

The useful test is concrete. An RTO needs evidence that the tool can support every required skill and a valid assessment judgement for the actual student cohort. A generic claim that an AI tool is innovative or efficient does not answer that question.

ASQA also reports seeing AI-drafted applications and non-compliance responses that use generic wording, retain drafting prompts or describe practices unsupported by evidence. The regulator's point is about accuracy and authenticity. AI-generated prose is not evidence that a process operates as described.

What this means for an RTO

There is no new commencement deadline or mandatory revalidation event from this publication alone. An RTO using AI should still be able to explain:

1. who approves and oversees each use;
2. where a person remains accountable for decisions;
3. how personal information and records are protected;
4. how unfair or inaccessible outcomes are identified and corrected; and
5. how AI-supported work remains consistent with the training product and assessment system.

If an RTO does not use AI in a particular workflow, RTObase should not infer that the guidance creates an action for that workflow.

How RTObase treats the publication

RTObase records this as verified editorial guidance. The 23 Outcome Standards controls remain pinned to the unchanged Outcome Standards instrument F2025L00354.

Existing assurance reviews do not become invalid merely because the guidance was published. Revalidation would be considered only if a reviewed future catalogue release changes a blueprint, source mapping or evidence contract.

This article is a general operational summary, not legal advice. RTOs should review their actual AI practices against ASQA's published guidance and obtain advice appropriate to their circumstances where needed.