In Australia there is unstoppable momentum toward AI-enabled WHS, with the strongest technical use cases in construction safety, hazard detection, computer vision, wearable monitoring, predictive risk analytics, robotics, digital twins and incident analysis. However, while the upsides are enormous, there are concerns and these relate to algorithmic management, worker surveillance, psychosocial risk, loss of autonomy, opacity, bias, privacy, weak consultation and unclear accountability.
For Boards and Executive Teams, it is important that AI in WHS should be governed as a material change to the system of work, not just a technology, privacy or cyber risk. This means directors need visibility over how AI changes work design, decision-making, control assurance, worker monitoring, learning, psychosocial risk and the allocation of accountability.
| Board governance issue | What the Board should require |
|---|---|
| 1. AI-WHS inventory | A register of all AI systems used in WHS or used to manage work, including safety analytics, monitoring tools, wearables, robotics, computer vision, scheduling tools, HR algorithms and generative AI tools used in safety administration. |
| 2. Risk classification | Classification of AI systems by potential WHS impact. Any AI that affects safety-critical decisions, worker monitoring, task allocation, fatigue, rostering, access control, performance assessment or control assurance should be treated as high governance interest. |
| 3. Safety case before deployment | A documented safety case for higher-risk AI tools. This should explain intended use, foreseeable misuse, hazards, data sources, validation, human oversight, fail-safe arrangements, limitations, consultation and monitoring. |
| 4. Worker consultation and representation | Evidence that workers and representatives have been consulted before deployment and after material changes. Consultation should cover purpose, data use, monitoring, intervention, appeal rights, psychosocial impacts and unintended consequences. |
| 5. Psychosocial risk assessment | Assessment of how AI affects job demands, job control, fairness, role clarity, surveillance, workload, fatigue, trust, under-reporting and work intensification. |
| 6. Procurement controls | AI procurement should require vendor disclosure on data, model limitations, validation evidence, cybersecurity, update processes, human oversight, audit access, incident notification and data retention. |
| 7. Independent assurance | Higher-risk AI systems should be subject to independent review, including algorithmic audit, technical validation, WHS assurance and legal/privacy review where appropriate. |
| 8. Change management | AI deployment should be treated as a change to the system of work, not just an IT implementation. Management of change should cover training, supervision, work design, consultation, incident response and operating procedures. |
| 9. Human oversight | The Board should require clear rules for when humans must review, override or stop AI-generated decisions. This is especially important where AI influences work allocation, risk ranking, fatigue decisions, exclusion from work or escalation of safety concerns. |
| 10. Data governance | Controls should cover data minimisation, biometric and health data, secondary use, worker access to data, data retention, cross-border transfer, cybersecurity and privacy. |
| 11. Metrics beyond productivity | Reporting should include WHS benefit, control effectiveness, false positives, false negatives, worker complaints, psychosocial indicators, under-reporting signals, intervention quality and unintended consequences. |
| 12. Board capability | Directors need enough AI literacy to challenge management on model risk, data quality, work design, assurance, worker impacts and vendor claims. |
| 13. Incident and near-miss learning | AI-related incidents should be reportable through the WHS system. This includes failed detections, incorrect recommendations, unsafe automation, worker harm from monitoring, algorithmic bias and automation-driven work intensification. |
