Online RTOs and Vulnerable Student Cohorts: Balancing AI Literacy With Safeguarding Under Standard 2.5

In high-risk qualifications involving vulnerable people, online RTOs face a sharp tension: AI literacy is now a valued graduate skill, yet AI also lets a student produce assessment responses that look competent without any genuine understanding of the safeguarding, ethical and practical content behind them. A graduate of a community services, aged care, disability support or early childhood qualification who has done this is not merely academically dishonest; they may be dangerous to the people they will work with. An analysis of the Standards framework and the practical assessment design that resolves the tension, for an audience of online providers, assessors and the students and clients who depend on getting this right. Specific provisions should be verified against the current Outcome Standards Instrument.

The Question Behind the Tension

A question put to ASQA at a recent sector webinar named one of the most acute tensions in contemporary VET delivery: in high-risk qualifications involving vulnerable people, how can online RTOs balance the responsible use of AI and the development of digital literacy with the need to verify genuine safeguarding knowledge, ethical judgement and competency? The framing is precise. It recognises that AI literacy is now a legitimate and valued component of professional education, while identifying the assessment integrity problem it creates when the qualification's purpose is to produce practitioners who will work with vulnerable people. A graduate who has used AI to produce assessment responses without genuinely understanding the safeguarding, ethical and practical content those responses describe is not merely academically dishonest. They are potentially dangerous to the people they will work with. The assessment system must verify genuine understanding, not merely the production of correct-looking responses. This article maps the Standards framework for that challenge and sets out practical assessment design for online RTOs delivering these qualifications.

1. The Compound Challenge: AI Literacy as Both Skill and Risk

Online RTOs delivering community services and health qualifications face a compound challenge that does not arise in most other contexts. AI literacy is increasingly recognised as a graduate attribute in these fields: practitioners who can use AI tools to navigate regulatory information, prepare documentation, access research and support case management are more effective than those who cannot, and employers increasingly expect that competency. At the same time, the assessment integrity challenge AI creates is most acute in exactly these qualifications, because the consequence of assessing a student as competent in disability support, aged care or child safety when they are not is direct and potentially severe harm to people who have little ability to protect themselves from an ill-equipped practitioner.

The challenge is therefore not a choice between AI literacy and safeguarding. It is an assessment design problem: how does the assessment system verify genuine understanding and ethical judgement in a way AI cannot replicate, while preparing students to use AI responsibly as a professional tool? These requirements are not in conflict; they call for different assessment designs for different competency domains. Standard 1.1 requires training modes and delivery to enable students to attain the skills and knowledge of the training product, and for AI literacy as a graduate skill, online delivery with AI-integrated learning activities is an appropriate modality. Standard 1.4 requires assessment to include practical application that the student must perform themselves. The two point in the same direction: teach AI use, but assess it, alongside genuine professional competency, in ways that require the student to demonstrate real understanding.

2. Standard 2.5 and the Safe Learning Environment Online

Standard 2.5 requires the RTO to foster a safe and inclusive learning environment for VET students, and a culturally safe learning environment for First Nations people. In online delivery of vulnerable-cohort qualifications, this operates differently from face-to-face delivery. In a classroom, trainers can observe engagement, notice early indicators of difficulty, read non-verbal cues and build relational rapport. Online, these observational mechanisms are largely absent, and a student struggling with content, distressed by case study material about abuse or neglect, or holding beliefs inconsistent with safeguarding obligations may simply not be visible to the trainer.

The safe learning environment obligation therefore needs specific design attention online. The environment should include clear channels for students to raise concerns, including about content that triggers a personal response; explicit warnings before material that may be distressing; wellbeing support resources accessible within the platform; trainer responsiveness expectations that are realistic for students who cannot approach a trainer after class; and assessment design that does not inadvertently create disclosure obligations for students who answer using their own personal experience of the content. This last point matters most for qualifications touching child protection, family violence, mental health and sexual assault, where tasks asking students to apply knowledge to personal scenarios may elicit disclosures that trigger mandatory reporting or require a trauma-informed response. A trainer who receives such a disclosure through a submission channel is not positioned to respond as a face-to-face trainer could, so the assessment design must be reviewed for inadvertent disclosure risk, and the trainer must have a documented protocol for responding when a disclosure arrives through an online channel.

3. Standard 1.4: What Genuine Safeguarding Knowledge Requires in Assessment

Standard 1.4 requires assessment to include practical application that lets the student demonstrate skills and knowledge in a practical setting. For safeguarding, ethical judgement and professional discretion, that means assessment must go beyond knowledge recall. A student who can state the legal definition of a reportable assault has demonstrated knowledge. A student who can identify a reportable incident in a contextualised case study, apply the relevant legal threshold, document the incident correctly and articulate the ethical obligations of their reporting role has demonstrated the practical application the Standard requires.

The problem with AI here is not that AI lacks the content. Current language models have been trained on large quantities of professional guidance, legislative frameworks and case material in community services, health and child protection, so a student who feeds a case study prompt to an AI can receive a response that correctly identifies the threshold, cites the legislation and describes the procedure. The response looks like competency. It is not. The distinction between an AI-generated correct-looking response and genuine competency lies in personalisation, contextualisation and professional reasoning. AI can reproduce the general principle; it cannot apply it to the specific, contextualised, personally experienced scenario that genuine competency requires. Tasks that require students to apply safeguarding knowledge to their specific placement context, to articulate their own professional reasoning, to identify the ethical tensions in a situation they have actually encountered, and to demonstrate understanding through oral questioning cannot be completed by AI on behalf of a student who has not genuinely engaged with the training.

4. The Standard 4.3(d) Child-Safe Requirement Online

Standard 4.3 addresses risk management, and Standard 4.3(d) requires that, where an RTO offers training or assessment to VET students aged under 18, risks to their safety and wellbeing are identified and managed by having regard to the training content and modes of delivery, and in accordance with the National Principles for Child Safe Organisations. This has specific implications for online RTOs delivering to school-based students or to cohorts that include young people under 18.

The National Principles for Child Safe Organisations call for a child-safe culture, policies and procedures addressing child safety, safe physical and online environments, strategies for identifying and responding to child safety concerns, a focus on empowering children and young people, and processes for responding to incidents and disclosures. For online RTOs, the online-environment dimension is directly engaged: the learning platform, communication channels, video tools and discussion forums must be configured and managed to provide a safe online environment for young students. The intersection with the safeguarding-knowledge challenge is acute, because a school-based student studying a community services qualification who is asked to respond to scenarios about child abuse or neglect is at once a learner engaging with professional content and a young person who may have personal vulnerabilities to that content. The assessment design must account for both. In practice, tasks involving sensitive safeguarding content should carry clear content warnings, offer alternative scenarios where possible for students who indicate a personal connection to the material, and sit behind clear trainer protocols for receiving and responding to disclosures. The obligation is not satisfied by a policy statement; it requires operational child-safe practices demonstrably applied in the online environment.

5. Assessment Design for Community Services Qualifications: A Domain Framework

The table below maps the key competency domains in community services and health qualifications against the appropriate role of AI in each, and the verification mechanism that satisfies the practical-application requirement without being substitutable by AI.

6. Online Assessment Design Principles for Vulnerable-Cohort Qualifications

Four principles, grounded in the practical-application requirements of Standard 1.4, address the online delivery challenge directly.

The first is to separate training activities from summative assessment. Online platforms make it easy to use AI on any text-based activity, so formative activities exploring content are an appropriate place for AI exploration, while summative tasks that determine competency must be designed with AI substitutability explicitly considered. The boundary between training and assessment, explored elsewhere in this series, must be operationally visible in the platform design.

The second is to prioritise placement-based evidence for safety-critical competencies. For safeguarding, mandatory reporting and emergency response, the most authenticity-robust evidence is workplace evidence from supervised placement. A student who applies mandatory reporting knowledge in a supervised workplace, with supervisor confirmation, has provided evidence AI cannot replicate. The system should weight placement-based evidence heavily for safety-critical domains and treat online written responses as supplementary knowledge evidence rather than primary competency evidence.

The third is to build oral verification into safety-critical tasks. Oral questioning remains the most reliable AI-resistant verification available online. A written case study response followed by a structured video oral examination covering the same material cannot have relied entirely on AI for the oral component. Video examination using a standard platform, with the recording retained in the assessment file, supports the authenticity requirement under Standard 1.4 while giving the assessor direct evidence of genuine understanding. The protocol should specify the examination's duration and scope, which tasks it supplements, a questioning framework focused on the student's reasoning rather than their conclusions, and the recording, retention and documentation requirements, because an undocumented oral examination produces no audit trail for the individual-judgement requirement.

The fourth is to use student-specific scenario design to prevent AI generalisation. Generic scenarios described in abstract terms can be answered by AI from general guidance. Scenarios that require the student to apply concepts to their specific placement, their supervisor's practice, a specific policy from their employer, or a specific incident they have personally encountered cannot be answered, because AI does not have access to that information.

7. A Pre-Deployment Review Framework

The checklist below consolidates these principles into a review to run on every summative assessment task before it is used for a competency determination.

Conclusion: Design, Not Detection

The instinct when AI threatens assessment integrity is to reach for detection: tools that flag AI-generated text, rules that ban AI use, surveillance of submissions. In vulnerable-cohort qualifications, that instinct is both unreliable and beside the point, because detection cannot tell the difference between a student who understands and a student who does not, only between text that looks AI-generated and text that does not. The durable answer is design. An assessment anchored in the student's own placement, their own documented interactions, their own real-time reasoning under oral questioning, and the confirmation of a workplace supervisor is one that AI cannot complete on the student's behalf, not because AI has been caught, but because the task requires something AI does not have. That same design teaches AI literacy where it belongs, in the professional toolkit, while reserving the competency judgement for evidence only a genuinely competent student can produce. For qualifications whose graduates will hold the safety and dignity of vulnerable people in their hands, that is not an academic nicety. It is the difference between a credential that protects those people and one that exposes them.

References and Further Reading

National Vocational Education and Training Regulator (Outcome Standards for Registered Training Organisations) Instrument 2025, including Standard 1.1 (training and delivery), Standard 1.4 (assessment and the rules of evidence), Standard 2.5 (safe and inclusive learning environment) and Standard 4.3 (risk management, including subparagraph (d) on students under 18). Federal Register of Legislation.

National Principles for Child Safe Organisations. National Office for Child Safety. https://www.childsafety.gov.au

Australian Skills Quality Authority. Practice Guides on diversity and inclusion, on assessment, and on risk management, and guidance on the non-compliant use of artificial intelligence. https://www.asqa.gov.au/rtos/2025-standards-rtos/practice-guides

Australian Skills Quality Authority. Standards for RTOs 2025 overview. https://www.asqa.gov.au/rtos/2025-standards-rtos