本文目前仅提供英文版本。
AI and Work Health & Safety in Australia: The Digital Work Systems Laws Every Employer Must Know
NSW became the first Australian jurisdiction to explicitly legislate WHS duties for AI and algorithmic systems in February 2026. Safe Work Australia is reviewing national model laws. Here is what Australian employers need to know now.
Key Takeaways
NSW passed the Work Health and Safety Amendment (Digital Work Systems) Act 2026 on 12 February 2026 — the first Australian jurisdiction to impose explicit WHS duties on employers using AI, algorithms, automation and online platforms in the workplace.
The primary duty of care now expressly includes ensuring workers health and safety is not put at risk from digital work systems — covering AI-driven work allocation, performance monitoring, rostering and surveillance tools.
PCBUs must specifically consider whether digital work systems create excessive or unreasonable workloads, unreasonable performance metrics, fatigue risks, psychosocial hazards, or discriminatory decision-making.
Safe Work Australia has been tasked with reviewing whether national model WHS laws should be updated to address the same subject matter. All Australian employers should prepare now regardless of jurisdiction.
WHS entry permit holders will gain new powers to access and inspect digital work systems relevant to a suspected WHS breach, subject to SafeWork NSW guidelines currently being developed following public consultation.
Psychosocial hazard regulations now apply in all Australian jurisdictions — Victoria completed implementation in December 2025. AI-driven monitoring, algorithmic performance management and work intensification all create cognisable psychosocial hazards.
"仅供参考。本文不构成法律、监管、财务或专业建议。如需具体指导，请咨询合格专家。"
The WHS framework and digital work systems
The Work Health and Safety Act (the model law, adopted in most jurisdictions) places on Persons Conducting a Business or Undertaking (PCBUs) a primary duty to ensure, so far as is reasonably practicable, the health and safety of workers. This duty applies to digital work systems including AI-powered tools, just as it applies to physical equipment and chemical hazards. The duty is non-delegable — you cannot outsource your WHS obligations to an AI vendor.
On 12 February 2026, New South Wales passed the Work Health and Safety Amendment (Digital Work Systems) Act 2026 — the first legislation in Australia that specifically addresses digital work systems in the WHS framework. The Act is awaiting proclamation. It signals the direction of Australian WHS law: AI-driven management systems, including productivity monitoring, algorithmic task allocation, and AI-generated workload management, are within the WHS regulatory framework and must be assessed for risk.
Psychosocial hazards: the key AI risk area
All Australian jurisdictions now have psychosocial hazard regulations requiring PCBUs to identify psychosocial hazards, assess the risk of psychological harm, implement controls, and monitor outcomes. Victoria was the last jurisdiction to commence its regulations — the OHS (Psychological Health) Regulations 2025 commenced 1 December 2025.
AI-related psychosocial hazards that employers must assess and control include:
Constant monitoring and surveillance: AI productivity tracking that measures every action an employee takes — keystrokes, mouse movements, application switches, idle time — creates an environment of constant surveillance. Research consistently shows this generates anxiety, reduced autonomy, and psychological harm even when the surveillance is disclosed. This is a psychosocial hazard that must be assessed and controlled.
Algorithmic performance management: AI systems that generate performance scores based on metrics the employee cannot see, contest, or predict create a form of power imbalance and uncertainty that the psychosocial hazard literature identifies as a source of psychological harm. Unpredictable, opaque performance management systems are a known psychosocial risk regardless of whether they are AI-driven.
AI-driven work intensification: AI scheduling systems that optimise throughput can set paces that are physically achievable but psychologically unsustainable. Amazon warehouse picking systems and Uber driver dispatch systems are documented examples. The AI optimises the aggregate; the individual worker bears the psychological cost of operating at the AI's optimal pace indefinitely.
Lack of control and autonomy: When an AI system allocates tasks, sets deadlines, monitors progress, and adjusts workload — with limited worker input — it reduces the worker's sense of control and autonomy. Loss of control over work is one of the strongest predictors of psychological harm in occupational health research.
Assessing and controlling AI-related WHS risks
PCBUs must apply the hierarchy of controls to AI-related WHS risks. Elimination — removing the hazardous aspect of the system — is the most effective control. If AI productivity monitoring creates surveillance anxiety, eliminating the surveillance feature is more effective than training workers to be resilient to it. Substitution — replacing a higher-risk approach with a lower-risk one — might mean using aggregate team metrics rather than individual tracking. Engineering controls — building in features that reduce risk, such as mandatory rest breaks that the AI cannot override, or transparency dashboards that let workers see and contest their AI-generated scores — are more effective than administrative controls alone. Administrative controls and training are important complements but cannot be the primary or only control measure under the new psychosocial regulations in Victoria (and the model approach elsewhere).
Document your risk assessment, the controls implemented, and the monitoring approach. Review at defined intervals and after any significant change to the AI system or following a reported incident or complaint.
Practical steps for employers
Audit your AI work systems: list every AI tool used in workforce management (scheduling, performance monitoring, task allocation, productivity tracking); for each, assess the psychosocial hazard profile; consult workers and their representatives about their experience — workers often identify hazards management cannot see from outside the workflow; implement controls proportionate to the risk level; monitor effectiveness; review annually and after significant changes. Ensure your psychosocial hazard risk assessment specifically addresses AI-related hazards. This is expected by regulators across all jurisdictions following the introduction of psychosocial regulations.