This is new guidance, not a new workplace AI ban
Safe Work Australia released a new set of guidance pages on 23 July 2026 explaining how work health and safety duties apply when businesses introduce or change artificial intelligence and other digital technologies. The central message is deliberately practical: technology can create opportunities, but it can also change physical and psychological risks at work.
The publication does not announce a separate AI licence, a list of prohibited tools or a new fine for using generative AI. It explains how the existing model WHS framework applies. States and territories regulate and enforce workplace safety laws, and Safe Work Australia says people needing advice about a specific workplace should contact the relevant regulator.
The scope is much wider than ChatGPT
The guidance uses a broad definition of digital technology. Its examples include computers, phones, applications, security systems, wearables, robotics and automation. AI includes predictive systems, machine learning and generative tools, but the safety questions are not limited to a worker typing prompts into a chatbot.
A rostering algorithm can affect fatigue. Monitoring software can reduce job control. An automated gate or collaborative robot can create a physical hazard. A customer-facing platform can expose workers to online abuse. Looking only for a humanoid robot would miss most of the systems the guidance is actually talking about.
A routine software update can still change the risk
Safe Work Australia specifically includes changes to technology already in use, even routine IT changes or updates pushed by a software provider. A familiar tool can begin allocating tasks differently, collecting more data or changing the pace of work without arriving in a new box.
That means a risk check should be tied to what the system does and how work changes, not only to the purchase date. If an update introduces a new automated decision, monitoring feature or workload target, the old assessment may no longer describe the job workers are now doing.
The duty begins with consultation, not a finished rollout
Under the model WHS approach, a person conducting a business or undertaking, usually shortened to PCBU, must consult workers and their health and safety representatives about matters that may affect health or safety. The new guidance says this includes consultation before introducing digital technology that may change those risks.
Consultation is more than announcing that a system will arrive on Monday. Workers can identify where a target is unrealistic, where an interface creates extra steps or where a monitoring rule changes breaks and customer interactions. The guidance also points to cooperation between businesses when more than one duty holder is involved in the same work.
The risk process still has four familiar stages
The guidance sets out the usual sequence: identify hazards, assess risks when they are not already known, control the risks, then monitor and review whether the controls work. A business must eliminate risks where reasonably practicable or minimise them so far as is reasonably practicable.
This is not a one-page AI policy that can be filed away after procurement. The assessment needs to connect a technology feature to real work: who uses it, which decisions it makes, what happens when it fails, how workers challenge an output and whether the system changes workload, movement, supervision or exposure to harmful behaviour.
Automation can remove easy work and intensify what remains
One of the guidance's most useful examples concerns work intensification. Automating routine tasks can sound like an obvious reduction in workload, but it may leave workers with a larger proportion of complex, emotionally demanding or cognitively tiring tasks. Fewer simple tasks do not automatically mean a safer day.
The opposite problem is possible too. Automating challenging parts of a role can make work repetitive and unstimulating, while constant checking of machine output can reduce autonomy. The safety question is not whether automation is good or bad in the abstract. It is how the whole job changes around it.
The less visible risks include surveillance, bias and disconnection
Safe Work Australia lists close monitoring, computer-set work pace, poor role clarity, inadequate training, unfair automated decisions and reduced human feedback among the possible psychosocial hazards. A system can be technically accurate and still be introduced in a way that reduces control or creates unreasonable demands.
Digital work can also blur the end of the working day. Constant connection may extend hours or interrupt sleep. Customer platforms and social media can increase anonymous abuse or harassment. The guidance treats those outcomes as work design and safety issues, not simply as a worker's failure to cope with technology.
Human oversight needs the power to change an outcome
The suggested controls include choosing tools designed with safety in mind, allowing workers to change pace or pause tasks, involving IT and WHS expertise, providing accessibility and moderation options, training people properly and auditing whether a system is working as intended.
A human reviewer is not a meaningful safeguard if that person lacks time, information or authority to correct the system. A practical plan identifies who can stop an unsafe workflow, override a faulty allocation, investigate a biased result and explain a decision to the worker affected by it.
Controls must be reviewed when the system or evidence changes
The guidance says control measures should be reviewed regularly and when they are not effective, a new hazard appears, consultation shows a review is needed or a change is likely to create a different risk. A software update can therefore be both a technology event and a safety review trigger.
Useful review evidence can include workload patterns, fatigue reports, near misses, complaints, override rates and whether workers can raise concerns without penalty. A dashboard claiming improved efficiency is not enough if the people using the system report that breaks have disappeared or complex work has accumulated.
What an Australian workplace can do before the next AI rollout
Start by writing down the work problem the tool is meant to solve, the decisions it will influence and the people whose work will change. Consult those workers before selecting or configuring the system. Identify physical and psychosocial hazards, choose controls, assign a person with authority to respond and set a review date before launch.
Workers should be told what the technology does, how it will be used, where its limits are and how to question an output or report a safety concern. Businesses should check the law and regulator guidance that applies in their jurisdiction. The new national material is a strong risk-management starting point, not personalised legal advice or proof that every use of AI is safe.
Sources and further reading
- Safe Work Australia - New AI and digital technologies guidance now available ↗
- Safe Work Australia - Artificial intelligence and digital technologies ↗
- Safe Work Australia - AI and digital technologies WHS duties ↗
- Safe Work Australia - Managing AI and digital technology risks ↗
- Safe Work Australia - Example opportunities and risks ↗
This article was written for Curiosity Desk. We do not copy other publishers or invent quotes. If a material error is found, we correct it openly.
Read the full standards →One answer should lead to a better question
Bring your curiosity to the group
Curious Minds is our public Facebook community for surprising science, strange history, Australian wildlife and everyday questions. No copied posts, no personal-friend invitations and no link dumping.
- Three self-contained discussion prompts each week
- Sourced answers and honest uncertainty
- Respectful conversation without spam
