Responsible AI · UB-AI-005
AI Human Oversight
A human trainer stays accountable. AI never makes a decision that affects you on its own.
AI can support a trainer. It cannot be the trainer, and it cannot be the person you appeal to.
What it covers
- AI-assisted course content, live AI explanations, learner support, debriefs, feedback and any AI-supported assessment or coaching.
- The optional participant AI assistant and the AI layer in our training platform.
- Any AI output that could materially influence your understanding, completion, assessment result or access to support.
What we commit to
- The human trainer remains accountable for learning outcomes, factual corrections, assessment decisions and escalated support.
- AI supports professional judgment; it does not replace it. You can always ask the trainer instead.
- In live sessions the lead trainer owns the learning flow while the support trainer monitors engagement and helps individuals.
- Any AI layer is optional and facilitator-controlled. The trainer can disable it at any time, and it is never the final authority on course content.
- Any AI-generated feedback, assessment, coaching or recommendation that could affect completion or another material decision requires human review before it is used.
- Material trainer interventions, overrides, corrections and escalations are logged. An absence of interventions is not treated as proof that oversight is unnecessary.
- Post-session queries that need review are acknowledged within two working days.
Escalation, in practice
- Trigger: you ask for help, an AI answer looks wrong or uncertain, output conflicts with the course material, something sensitive or out of scope comes up, or a decision could be affected.
- Human check: the trainer goes back to the source or course standard and decides whether the output can be clarified, corrected or has to be withdrawn.
- Intervene: the trainer gives the human answer, corrects the record, pauses or disables the AI feature if needed, and stops reliance on the flawed output.
- Escalate: privacy, safeguarding, legal uncertainty or repeated system failure goes to the relevant owner and, where appropriate, the client contact.
- Record and improve: the intervention is logged, materials, prompts, restrictions or settings are updated, and it closes only when follow-up is complete.