When the Shift Plan Is an Algorithm: NSW Raises the Evidence Bar for Digital Work Systems

MineGuard AI

27 October 2026 7 min read

A mine can change how work is allocated without changing a roster template, a dispatch screen or a supervisor’s title. The change may sit inside an algorithm that sequences tasks, sets priorities, tracks performance or recommends who does what next.

That makes the latest New South Wales work health and safety development more than an employment-law issue. It is an operational-governance issue for mining leaders.

The Work Health and Safety Amendment (Digital Work Systems) Act 2026 defines a digital work system broadly as an algorithm, artificial intelligence, automation or online platform. Among other changes, it will clarify that a PCBU’s primary duty extends to risks arising from digital work systems and introduce a specific duty concerning the allocation of work by those systems.

Most operative provisions have not yet commenced. SafeWork NSW says only the provisions enabling development of the Digital Work Systems Guidelines have started; the remaining provisions await proclamation. The new entry-permit-holder power cannot commence earlier than one month after the Guidelines are published.

The timing still matters. SafeWork NSW has completed Stage 1 consultation, published its October 2026 insights report, and plans further targeted consultation in late 2026 before broader public consultation on draft Guidelines.

For executives, the signal is clear: if a digital system influences work, the organisation should be able to explain its safety impact in operational evidence, not vendor language.

Why this belongs on the mine’s risk agenda

The general NSW WHS framework applies to mines alongside mine-specific legislation, as the NSW Resources Regulator confirms.

On a mine, digital work systems may include rostering and fatigue tools, fleet or task-dispatch systems, maintenance work queues, contractor platforms, worker-monitoring technologies, automated plant and AI-enabled safety workflows. Not every system carries the same risk. The useful question is not “does it contain AI?” but:

How much influence does this system have over who performs work, when, at what pace, with what information, and with what opportunity for a person to intervene?

That distinction is reflected in SafeWork NSW’s consultation findings. Stakeholders described systems ranging from administrative tools to highly automated systems that make operational decisions with limited human involvement. They also emphasised that risk depends on how a system is designed, configured, implemented and used, not only on the technology itself.

Safe Work Australia’s 2026 guidance on AI and digital technologies identifies potential physical and psychosocial risks including work intensification, fatigue, reduced job control, unreasonable monitoring, poor change management and automated allocation without adequate human oversight. It also says controls should be reviewed before a software change likely to create new or different WHS risks.

A system can therefore be technically available, contractually supported and operationally accepted while its safety case remains weak.

The evidence bar is moving from policy to inspectability

The pending reforms do not create an unrestricted right to inspect every system. SafeWork NSW states that access must be relevant to a suspected contravention and remains subject to existing WHS Act conditions and the future Guidelines.

There is an important practical distinction:

Entry permit holders already have a right, subject to the Act, to inspect a work system relevant to a suspected contravention, which can include a digital work system.
The new, uncommenced power would allow them to require reasonable assistance from a PCBU to access and inspect a relevant digital work system.
The Guidelines will shape how that power operates and may apply only to specified classes of persons, work or workplaces.

SafeWork NSW’s Stage 1 report shows what “understanding the system” may involve in practice. Stakeholders discussed explanations, demonstrations, risk assessments, reports, activity logs, configuration records, change logs, business-decision records and access to people with relevant technical expertise. They also raised the difficulty of systems owned by vendors, parent companies or offshore providers.

That is the leadership issue. Outsourcing the software does not outsource the need to understand how it shapes work or how its risks are controlled.

Five things mining leaders should commission now

1. Build a digital work-system register

Start with systems that allocate, pace, monitor or assess work, or influence a safety-critical decision. Record the owner, purpose, users, affected workers, vendor, data sources, integrations, automated actions and current version.

A software asset list is not enough. The register should show each system’s influence on work and the possible physical or psychosocial consequences if it is wrong, unavailable, misconfigured or used outside its intended boundary.

2. Map the decision chain

For each material system, document:

what enters the system
what rule, model or configuration transforms the input
what output reaches the worker or supervisor
whether the output is advisory or directive
who can question, override or stop it
what happens when data is missing or conditions change
where the decision and intervention are logged

This makes human accountability testable. A nominal “human in the loop” is weak if the person lacks time, information, authority or a workable override.

3. Test the work, not only the software

Acceptance testing should include the actual operating context: shift pressure, degraded communications, competing production priorities, contractor interfaces, fatigue, unusual tasks and software updates.

Consult the people whose work is allocated or monitored. Safe Work Australia’s guidance requires the familiar WHS cycle of identifying hazards, assessing and controlling risks, then monitoring and reviewing controls in consultation with workers and their representatives. Worker experience is evidence about system performance, not a late-stage adoption exercise.

4. Preserve an inspection-ready evidence pack

For higher-consequence systems, maintain a concise, current pack containing:

approved purpose and prohibited uses
accountable business, safety and technical owners
risk assessment and consultation records
configuration and version history
test criteria, results and known limitations
relevant inputs, outputs, alerts and overrides
incidents, complaints and corrective actions
vendor responsibilities and escalation paths
retention, access and privacy rules
change approvals and reassessment triggers

The aim is not to predict the final Guidelines. It is to ensure the organisation can reconstruct how the system operated, what it influenced and who accepted the controls at the relevant time.

5. Put vendor access and change control into the contract

The Stage 1 report identifies a practical problem: relevant functionality, records or expertise may sit with a third-party provider. Procurement should therefore address access to logs and configuration information, notification of material changes, support for investigations, named technical contacts, retention periods and exit arrangements.

If the operator cannot obtain enough evidence to understand a system’s safety influence, that limitation belongs in the risk decision before deployment.

What this changes for executive assurance

Boards and senior leaders do not need source code reviews for every workplace tool. They do need assurance proportionate to the influence the system has over work.

A useful executive view should show:

which systems can materially shape work allocation or safety decisions
where the highest physical and psychosocial exposures sit
whether decision boundaries and human interventions are real
whether worker consultation changed the design or controls
whether third-party dependencies are understood
whether material updates trigger reassessment
whether the organisation can reconstruct a decision after an incident or complaint

This is the practical bridge between WHS governance and AI governance. It moves the conversation from “we have an AI policy” to “we can show how this system behaved in this workflow, on this version, with these controls.”

How MineGuard AI can help

This problem maps directly to MineGuard AI’s documented consulting services, without asking software to replace accountable judgement.

MineGuard AI Consulting covers AI governance and strategy, data readiness, safety-workflow transformation, and implementation and integration. Its published deliverables include governance playbooks, RACI models, assurance checklists, metadata and ingestion rules, future-state workflows, integration designs and deployment runbooks. The service also addresses human decision boundaries, logging requirements, versioning, approvals and change control.

MineGuard’s own Trust & Security materials describe the operating principles applied to its products: AI outputs are presented for human inspection, named approval is required before material reaches the field, and approval activity is logged. Those product controls do not discharge a customer’s WHS duties, but they illustrate the evidence-anchored, human-accountable approach required in high-consequence work.

A sensible first engagement is an AI Adoption & Governance Sprint focused on one influential digital work system. Map the real workflow, identify the affected workers and decision boundaries, test the evidence available from the vendor and build an assurance pack that can survive operational, workforce and regulatory scrutiny.

Talk to MineGuard AI about making a digital work system governable before its authority outruns its evidence.

Key sources

This article provides general information, not legal advice. Confirm obligations and commencement dates for your operation with the relevant regulator and qualified advisers.