AI Transparency Statement

1. Why This Statement Exists

In preparation for obligations commencing under the Privacy and Other Legislation Amendment Act 2024 (Cth) (Schedule 2, effective approximately June 2026), High Risk Interactive proactively discloses how it uses substantially automated decision-making that may significantly affect individuals. This statement explains how we use artificial intelligence (AI) in the Platform and what safeguards are in place.

We believe transparency about AI use is essential in vocational education, where assessment decisions can affect a learner's career, qualifications, and ability to work in high-risk industries.

2. AI Systems We Use

The Platform uses AI in the following areas:

2.1 Assessment Marking

What it does: Compares student assessment responses against the pre-set correct answers, marking rubrics, and model answers defined by the assessor in the assessment tool, and provides a suggested competency outcome (Competent / Not Yet Competent) with feedback.

Two marking modes are available:

Option A — Local Marking (Default, Recommended)

Student responses are compared against the assessor's set answers entirely on our own servers using keyword matching, concept coverage analysis, and VET/WHS synonym expansion. No student data is sent to any external provider. This mode:

• Extracts key terms from the assessor's model answer
• Checks how many appear in the student's response (with typo tolerance)
• Expands VET industry abbreviations (e.g., PPE = personal protective equipment, SWL = safe working load)
• Scores concept coverage sentence-by-sentence
• Generates deterministic, reproducible feedback

Option B — AI-Assisted Marking (Optional)

If enabled by the RTO administrator, student responses may be sent to our AI provider (Anthropic Claude API) along with the assessor's pre-defined correct answers for more nuanced comparison. This is useful for longer essay-style responses where semantic understanding adds value.

Key point: In both modes, the system does not independently judge competency. It measures the student's response against answers that a qualified assessor has already defined. The set answers are the source of truth.

Human oversight: All marking outcomes are flagged for human assessor review when confidence is below the configured threshold. A qualified human assessor must confirm, modify, or override the suggestion before it becomes the final result. Marking is a decision-support tool, not the decision-maker.

2.2 SkillMate AI Tutors & Study Assistant

What it does: Provides on-demand tutoring conversations where learners can ask questions about course content, get explanations, and practice knowledge. Enrolled learners also get the HRI Study Assistant — a cross-course study helper with daily and monthly usage limits.

How it works: Each SkillMate bot (DogMate, CraneMate, ForkMate, HeightsMate, RigMate, PlatformMate, SafetyTutor) has a hardcoded knowledge base sourced from Australian Standards, Codes of Practice, WHS Acts, and unit of competency requirements. The AI responds based on this verified knowledge.

Safeguards: Tutors have absolute rules that cannot be overridden — they will not provide information outside their knowledge base, will not guess at safety-critical information, and always caveat answers with "check manufacturer specifications" and "refer to your trainer." Tutors never mark or grade anything.

2.3 AI Question Generation

What it does: Generates quiz questions and assessment items aligned to specific units of competency.

How it works: The AI generates questions based on the unit's elements, performance criteria, and knowledge evidence requirements. Generated questions are reviewed and approved by a qualified trainer/assessor before being used in assessments.

Human oversight: No AI-generated question is presented to learners without human review and approval.

2.4 AI Content Analysis

What it does: Analyses uploaded documents, checks plagiarism indicators, and assists with Recognition of Prior Learning (RPL) evidence assessment.

How it works: Document content is sent to the AI provider for analysis. Results are presented to the assessor as advisory information.

Human oversight: All RPL and plagiarism decisions are made by qualified human assessors. AI analysis is one input among many.

2.5 Oral Practice Assistant

What it does: An adaptive oral-questioning assistant that helps probe understanding of course topics and produces evidence and a recommendation for the assessor.

Human oversight: The assistant is explicitly not the assessor. It never makes the competency determination — a qualified human assessor reviews its evidence and makes every determination of record.

2.6 Content & Media Generation (Staff-Facing)

What it does: Our staff use AI tools to produce course and marketing media — image generation, video generation, and text-to-speech narration (including a self-hosted voice engine that runs entirely on our own hardware).

What data is involved: Text prompts and course material only — no learner data is used in media generation.

Human oversight: Staff review every generated image, video and narration before it appears in any course or marketing material, and all safety-related content is verified against registered Australian Standards and Codes of Practice sources before publication.

2.7 Administrative AI

What it does: AI suggests draft replies in our support and enquiry inbox, and powers a phone demonstration agent for prospective clients.

Human oversight: AI never sends a message — a staff member reviews and sends every reply. Phone callers are told at the start of the call that they are speaking with an AI agent.

3. What Data Is Processed by AI

3.1 Local Marking (Default)

When local marking is enabled, no assessment data is sent to any external provider. All comparison processing happens on our own servers. Student responses are only compared against the assessor's set answers locally.

3.2 AI-Assisted Features (When Enabled)

When AI features are used, the following data may be sent to our AI providers:

• Written assessment response text (only if AI marking mode is enabled by the RTO administrator)
• Tutor conversation messages (questions you ask and context)
• Document text content (for plagiarism checking or RPL analysis)
• Course content text (for question generation)

What is NOT sent:

• Your name, email, or other identifying information (unless included in your assessment response text)
• USI numbers, disability status, Indigenous status, or visa information
• Payment or financial information
• Passwords or authentication credentials

4. AI Providers

Different features use different providers. Learner-facing tutoring and AI-assisted marking run on Anthropic Claude. Staff-facing drafting and generation features use a provider chain led by Google Gemini. Narration uses our own self-hosted voice engine.

Anthropic (Claude API) — Tutors, Study Assistant & AI-Assisted Marking

• Location: United States
• Data retention: API inputs are not retained for model training
• Processing: Data is processed in real-time and not stored beyond the session

Google (Gemini API) — Content Drafting, Image & Video Generation

• Location: United States
• Used for staff-facing content drafting, question generation, and media generation
• No learner personal information is sent through these features — course material and prompts only

OpenAI — Fallback

• Location: United States
• Data retention: Business API terms — inputs not used for training
• Configured as a fallback only and currently disabled by default

Retell AI + ElevenLabs — Phone Demonstration Agent

• Used only for the opt-in phone demo experience; callers are told they are speaking with an AI
• Call logging is disabled

HRI Voice — Self-Hosted Narration

• Runs entirely on High Risk Interactive's own hardware — no data leaves our systems

Every AI system we use is screened before adoption and recorded on our internal AI Systems Register with a named accountable owner, the data classes it may process, its human-oversight point, and a review date. For sub-processor details, see our Sub-Processor Register.

5. Human Oversight Guarantees

We guarantee that:

• No competency outcome is determined solely by AI. A qualified human assessor always makes the final decision.
• No certificate or statement of attainment is issued based solely on AI assessment.
• No learner is disadvantaged by AI errors — if an AI suggestion is incorrect, the human assessor corrects it.
• Learners can request that their assessment be reviewed entirely by a human assessor without AI involvement.
• AI tutor responses are clearly identified as AI-generated and come with appropriate caveats.
• Simulations and AI practice tools are training and familiarisation aids only — they are never used as competency assessment for high-risk work units.
• No high-risk work licence outcome is ever determined by AI. High-risk work assessment happens in person, on real plant, under SafeWork requirements.

6. Your Rights

Under the Privacy Act 1988 and our Platform policies, you have the right to:

• Know when AI is used: AI-assisted features are clearly labelled in the Platform interface
• Request human-only assessment: You may request that your assessments be marked entirely by a human assessor
• Appeal AI-influenced decisions: If you believe an AI-assisted assessment outcome is incorrect, you can request a reassessment
• Understand the logic: You may request an explanation of how AI was used in any assessment decision that affects you
• Access your data: You may request access to any data that was processed by AI on your behalf

To exercise any of these rights, contact your RTO administrator or email us directly.

7. Alignment with Australian AI Standards

Our AI practices are aligned with the Voluntary AI Safety Standard and the National AI Centre's Guidance for AI Adoption (Department of Industry, Science and Resources), and follow practices consistent with AS ISO/IEC 42001:2023 (AI management systems — we are aligned to, not certified against, that standard). We maintain the governance artifacts recommended for Australian training organisations — an AI policy, an AI screening process for new tools, an AI systems register, and this transparency statement. Our principles:

• Transparency: This statement and in-product labelling
• Accountability: Human assessors are accountable for all competency decisions
• Fairness: AI is applied consistently; outcomes are reviewed for bias
• Privacy: Minimal data sent to AI providers; no identifying information where possible
• Safety: AI tutors cannot override safety rules; hardcoded knowledge prevents hallucination on safety-critical topics
• Contestability: Learners can challenge any AI-influenced decision

8. Contact

For questions about our use of AI: