Does WHS law already cover AI work-allocation systems?
Yes. There is a common assumption that the law has not caught up with workplace AI. For psychosocial risk in New South Wales, that assumption is wrong. The duties that apply to an algorithmic rostering tool, an automated task-distribution engine, or a productivity-monitoring platform are the same duties that have applied to work design for years.
The starting point is the primary duty of care. Under section 19 of the Work Health and Safety Act 2011 (NSW), a person conducting a business or undertaking (PCBU) must ensure, so far as is reasonably practicable, the health and safety of workers. The Act defines "health" to include psychological health, and section 19(3) expressly requires the provision and maintenance of a work environment, and safe systems of work, that are without risks to that health. An AI tool that decides who does what, how fast, and under what level of observation is a system of work. It sits squarely inside the duty.
The model WHS framework, adopted in NSW, then makes the psychosocial obligation explicit. Following amendments that commenced on 1 October 2022 and were carried into the remade Work Health and Safety Regulation 2025 (NSW), clause 55A defines a psychosocial hazard, clause 55B defines a psychosocial risk, clause 55C requires the PCBU to manage that risk, and clause 55D sets out the matters a PCBU must consider when choosing control measures. None of these provisions mention AI, and none need to. They turn on the design and management of work, which is exactly what an allocation algorithm performs.
How do AI allocation and monitoring tools create psychosocial hazards?
Safe Work Australia and SafeWork NSW identify a recognised set of psychosocial hazards. Several map directly onto how automated work systems behave.
Low job control. Low job control means workers have little say over how or when work is done. When an algorithm assigns tasks, sequences them, and resets the queue without worker input, control shifts from the worker to the system. The worker becomes a recipient of instructions rather than a participant in planning.
High job demands and pace-setting. Job demands become a hazard when sustained effort outstrips capacity. Systems that optimise for throughput can ratchet up pace continuously, removing the natural recovery pauses that human-led scheduling leaves in place. Pace set by a metric, not a person, tends to drift toward the maximum the data will tolerate rather than the level a worker can sustain.
Poor role clarity. When allocation logic is opaque, workers may not understand why they received a task, what standard applies, or who is accountable for a decision the system made. Lack of role clarity is a listed hazard, and automated routing can erode it without anyone intending to.
Monitoring and surveillance. Continuous tracking of keystrokes, location, idle time or output is itself a psychosocial stressor. The sense of being constantly measured affects autonomy and trust even when no adverse action follows. These hazards rarely act alone. The Regulation specifically requires a PCBU to consider how psychosocial hazards may interact or combine, and a single platform can deliver low control, high demand and surveillance simultaneously, compounding the risk.
What does the SafeWork NSW Code of Practice require?
The Code of Practice: Managing psychosocial hazards at work (SafeWork NSW, May 2021) is an approved code under section 274 of the WHS Act. An approved code is admissible in proceedings as evidence of what is known about a hazard and reasonably practicable ways to manage it. A court can have regard to it when deciding whether a duty was met. In practice, following the Code is the most defensible path, and departing from it puts the onus on you to show your alternative was at least as effective.
The Code sets out a four-step risk management process: identify hazards, assess risks, control risks using the hierarchy of controls, and review the controls. It is explicit that control measures must follow the hierarchy under the Regulation, prioritising higher-order controls such as redesigning the work over lower-order controls such as training individuals to cope. For an AI allocation tool, that ordering matters. Adjusting the system so it cannot set an unsafe pace is a higher-order control. Offering a wellbeing webinar to staff worn down by that pace is a lower-order control and, on its own, unlikely to discharge the duty.
The Regulation reinforces this. Clause 36 sets the hierarchy of control measures, clause 55C requires the hierarchy to be applied to psychosocial risk, and clause 55D lists the matters you must weigh: the duration, frequency and severity of exposure; how hazards interact; the design of work, including job demands and tasks; and the systems of work, including how work is managed, organised and supported. An allocation algorithm touches every one of those matters.
What is the consultation duty, and when does it apply?
Consultation is not optional and not a courtesy. Under section 47 of the WHS Act, a PCBU must, so far as is reasonably practicable, consult with workers who are or are likely to be directly affected by a health and safety matter. Section 48 sets out what genuine consultation involves, including sharing relevant information, giving workers a reasonable opportunity to express views, and taking those views into account. Section 49 lists when consultation is required, and it captures the lifecycle of an AI rollout precisely: when identifying hazards and assessing risk, when making decisions about how to eliminate or minimise risk, when proposing changes that may affect health or safety, and when making decisions about procedures for monitoring conditions.
SafeWork NSW is direct on this point, describing consultation as "a legal requirement under the Work Health and Safety Act 2011" and "more than talking with your workers." It is a two-way process. Procuring or configuring a work-allocation system, then announcing it, is not consultation. The duty arises before the decision is locked in, while worker views can still shape the design, the metrics and the safeguards.
What to never do: do not treat a vendor demonstration or an IT procurement sign-off as the consultation step. Workers who will be allocated work, paced or monitored by the tool are the affected workers under section 47, and they, or their health and safety representatives, must be consulted on the WHS implications before deployment.
A step-by-step risk assessment for an AI work-allocation tool
The following workflow applies the four-step Code process to an automated allocation, pace-setting or monitoring system. It is functional guidance, not legal advice, and should be adapted to your workplace.
Step 1: Identify the hazards. Map what the tool actually does. Does it assign tasks, sequence them, set targets, measure output, or track activity? For each function, ask which psychosocial hazards it could introduce: low job control, high demands, poor role clarity, surveillance, or poor organisational justice if it drives decisions about people. Consult workers here under section 49; they will see effects that a specification document does not.
Step 2: Assess the risk. Using the clause 55D matters, evaluate the duration, frequency and severity of exposure. Is the system always on, or used occasionally? Can it set pace continuously, or within bounded limits? Assess how hazards combine. A tool that allocates work, scores performance and monitors activity carries cumulative risk that exceeds any single feature.
Step 3: Control the risk, highest order first. Eliminate hazards where reasonably practicable. Where elimination is not practicable, minimise risk by redesigning the system: cap maximum pace, build in recovery time, give workers the ability to query or adjust allocations, make the logic explainable, and constrain surveillance to what is genuinely necessary. Treat training and support as supplements, never as the primary control. Document why each control was chosen.
Step 4: Review. Clause 55C and the Code require controls to be maintained and reviewed. Monitor leading indicators such as workload complaints, turnover and absenteeism, and review after any change to the algorithm, because a model update can silently alter the work design you assessed. Re-consult when you review.
What about the NSW Digital Work Systems reforms?
NSW has legislated specifically in this area. Parliament passed an amendment to the WHS Act introducing an express duty directed at digital work systems, defined as an algorithm, artificial intelligence, automation or online platform, with stated considerations including excessive or unreasonable workloads, performance metrics, and monitoring and surveillance. This is a significant signal of regulatory direction.
It is important to be precise about status. As at the date of writing, the operative provisions of that amendment have not commenced; they are to begin on a day to be appointed by proclamation, which has not yet occurred. For that reason, this article does not cite the new section as operative law. The practical point for WHS teams is reassuring: you do not need to wait for proclamation. Every psychosocial hazard the new duty targets is already captured by the existing framework under sections 19 and 47 of the WHS Act and clauses 55A to 55D of the WHS Regulation. Teams that build their AI governance on the duties already in force will be well placed when the additional duty commences, rather than scrambling to catch up.
TheAICommand. Intelligence, At Your Command.
