Work health and safety

Only within limits. WorkSafe Victoria's code limits multilingual facilitators to general, simple and low-risk matters and requires employers to check the facilitator is confident and willing and that employees consent. Safe Work Australia notes a bilingual worker supporting others may need their workload modified. Use an accredited professional translator for complex or high-consequence content. From AI Translation Can Multiply a Safety Error Across Every Language.

No. The forum can identify a collision and assign an owner, but it cannot consult for affected workers or their HSRs. Under the WHS Act, sections 47 to 49 require consultation, so far as is reasonably practicable, with workers directly affected or likely to be directly affected, and the consultation must involve the health and safety representative where workers are represented. Use the map to route consultation, not compress it. From One AI Rollout Is Change. Ten Overlapping Updates Are an Exposure Pattern.

No. The support pack should describe the response, not profile workers. Comcare's guidance says it is not a manager's role to diagnose a mental health condition or to be a counsellor, and silence, camera status, leave use or sentiment scores must not be read as evidence of a person's mental state. From The First Manager Message After an Incident Can Help or Harm. Govern the Draft.

No. AI may assist with three tasks only: identifying candidate hazards, synthesising de-identified survey data into themes, and drafting the written assessment with rating fields left blank. Under Australian WHS law a competent person must determine the risk rating, select the controls, and sign off. The model never rates or signs. From AI-Assisted Psychosocial Risk Assessment: A WHS Governance Workflow That Keeps the Sign-Off Human

Yes, as a drafting tool using verified sources. AI can structure objectives, scenarios, questions, facilitator notes and plain-language explanations. A competent person must check the content against the current work, risks, controls, equipment, jurisdiction and workforce before delivery. From AI Can Draft WHS Training. It Cannot Verify Competence.

No. The incident may reveal a control, supervision, equipment or system-of-work failure rather than a knowledge gap. A competent person investigates, consults workers, reviews controls and decides what training, instruction, supervision or other action is required. From AI Can Draft WHS Training. It Cannot Verify Competence.

No. AI must never auto-classify whether an event is notifiable under sections 35 to 39 of the model WHS Act. The notification duty is immediate and strict, so an automated recommendation can reliably produce a breach. The safe pattern is the inverse: AI prepares the facts so a named competent person decides faster, leaving the decision field blank. From AI for Incident Analysis and Leading Indicators: A Human-in-the-Loop WHS Playbook

No. Notifiability is a statutory judgement the accountable person makes, with the regulator available for advice when in doubt. A well-designed skill structures the facts and presents the notifiable incident questions as a checklist for the human. It never outputs notifiable or not notifiable as a conclusion, a recommendation, or a probability. From An Incident Triage Skill File That Never Decides Notifiability

No. AI can compare supplied text and flag gaps for review, but it lacks the full workplace context and cannot make a legal or professional determination. A competent person must assess the plan against the current law, the hazards, the workforce, the location and regulator guidance. From AI Can Stress-Test an Emergency Plan. It Cannot Run the Drill.

AI can draft the consultation plan, structure the risk assessment, and turn workshop notes into a tidy first draft. It cannot consult, because consultation is a two-way exchange with the people who do the work. It cannot set the risk rating or decide which controls are reasonably practicable, and it cannot sign off or carry the accountability. Those stay with a competent person. From Rolling Out AI Is a Workplace Change: Consult and Risk-Assess First

Yes, for drafting only. The skill file works from de-identified learnings, keeps to the facts supplied, and stamps every output as a draft. The HSR is consulted and a competent person reviews and approves before anything is released to workers. The AI never verifies facts, never sets controls and never releases a communication. From The Safety Comms Skill File: Alerts Workers Actually Read

Not for the worker-facing OHS translation itself. WorkSafe Victoria's compliance code says machine-automated interpreting and translating tools cannot be guaranteed to be accurate and should not be used to translate OHS information. The code is not mandatory, but a duty holder who complies with it is considered to have complied to the extent it deals with their duties. AI may still help lock sources, compare versions or prepare a glossary. From AI Translation Can Multiply a Safety Error Across Every Language.

No. AI can produce a formatted first draft, but a SWMS for high-risk construction work must take account of the actual workplace under regulation 299(3) and be built in consultation with the workers who will do the work. A model has never seen your site and cannot hold the toolbox talk, so the draft is not yet lawful until a competent person makes it true. From AI Can Draft a SWMS in Seconds. The Site Walk and the Consultation Cannot Be Automated.

Yes. A camera trained on a person, a cab-facing fatigue camera, or a body-tracking wearable is a form of monitoring, and Comcare lists intrusive surveillance among the psychosocial hazards a PCBU must manage under regulations 55A to 55D. Because health in the WHS Act means physical and psychological health, the same primary duty covers it, supported by consultation. From An AI Safety Control Is Still Just a Control. The Duty to Verify It Stays With You.

Start with text and minimise collection. WorkSafe Victoria says an inspection need not be a physical visit because virtual viewing, video or photographs may be used, but home images can reveal household members, possessions and location details. Explain any additional collection, use an approved access-controlled system with retention rules, and offer a non-image route where practicable. From A Green Home-Office Score Does Not Clear the Risk.

Yes. Under the harmonised WHS framework an HSR can represent the work group, monitor the measures taken by the business, investigate complaints and inquire into apparent risks, and the PCBU must consult the HSR, confer when reasonably requested and allow access to information it has about relevant hazards, risks and worker health and safety. Consultation does not finish at deployment, and regulation 38 supports a control-review request when a review circumstance has not been adequately addressed. From An HSR Asked How the AI Tool Changes Work. A Vendor Slide Is Not an Answer.

Not as the current licence standard. The superseded Version 3 guide from September 2014 remains useful only as historical guidance on evidence-sufficiency technique. It described objective evidence as potentially including documents, electronic information, documented records, visual observations and discussion with workers and others, and it warned that conformance to its criteria did not assure compliance with every statutory obligation. From An AI-Written WHS Record Is Not Audit Evidence Until You Prove the Work Happened.

Yes, where the tool changes how people are allocated work, measured, or monitored. The consultation is not a launch email. It is a genuine opportunity for affected workers and their health and safety representatives to influence the decision before it is made, as required by the model framework consultation provisions in sections 47 to 49. From Rolling Out AI Is a Workplace Change: Consult and Risk-Assess First

Yes. Safe Work Australia's July 2026 AI guidance says implementing new digital technologies or changing existing ones, including routine changes to IT systems and updates pushed or initiated by software providers, may introduce or change WHS risks. Its examples identify workloads increasing to get across new system updates, particularly where updates are unnecessary, too frequent, or have little safety or long-term efficiency benefit. A supplier's minor release label does not answer the WHS question. From One AI Rollout Is Change. Ten Overlapping Updates Are an Exposure Pattern.

Yes, through the general psychosocial provisions. Regulations 55A to 55D of the Work Health and Safety Regulations 2011 (Cth), in Compilation No. 28, address psychosocial hazards and risks. Regulation 55D requires a PCBU selecting controls to have regard to all relevant matters, including exposure duration, frequency and severity, interacting hazards, work design including job demands, and systems of work. Other model-law jurisdictions may vary, so check the local instrument. From AI Made the Queue Faster. It Also Made the Job Harder.

Yes, so far as is reasonably practicable. Sections 47 to 49 of the WHS Act require consultation with directly affected workers when identifying hazards, deciding on controls, and proposing changes affecting health or safety. Consultation must occur before the decision is finalised, not after deployment, so worker views can shape the design. From AI Work Allocation and Psychosocial Risk: The WHS Duty NSW Teams Already Hold

No. Comcare's contractor-management guidance for Commonwealth PCBUs says a PCBU can rely on specialist expertise, but that reliance does not absolve the PCBU of its duties, and the legislation was not intended to create a general obligation to supervise a specialist contractor. Reasonable action depends on actual control, knowledge, expertise and capability, so target the interfaces you can affect. From The Contractor Used AI to Write the Safety Pack. Your WHS Duty Did Not Move.

No. The legal anchor is psychosocial risk management. Under the Commonwealth WHS Regulations, regulation 55C requires psychosocial risks to be managed and regulation 55D lists relevant control matters including systems of work and the information, training, instruction and supervision provided. The applicable code and regulations differ by jurisdiction, so check the framework that covers the workplace. From The First Manager Message After an Incident Can Help or Harm. Govern the Draft.

No. Under the Work Health and Safety Act 2011, section 14 states that a duty cannot be transferred and section 19 places the primary duty on the PCBU. A health and safety committee recommendation can inform a control decision, but the decision and the duty remain with the PCBU acting through authorised management. From Your WHS Committee Pack Should Ask for Decisions, Not Bury Them in Updates.

No. Section 14 of the Commonwealth WHS Act says a duty cannot be transferred, and section 16 allows more than one person to hold the same duty while preserving each person's responsibility according to their capacity to influence and control the matter. Authorship of the document, human or AI, does not change who holds the duty. From The Contractor Used AI to Write the Safety Pack. Your WHS Duty Did Not Move.

A tabletop can test decision logic and expose assumptions, but it may not test physical evacuation, alarm audibility, equipment, access or worker behaviour. The appropriate testing program depends on the workplace and is set by a competent person using current jurisdictional guidance, and regulation 43 requires the plan to state the frequency of testing. From AI Can Stress-Test an Emergency Plan. It Cannot Run the Drill.

No. The OAIC's guidance says the use and disclosure of personal information must comply with APP 6. An approved tool, a WHS purpose label and removing direct identifiers do not themselves authorise a secondary use. Confirm the APP 6 basis with privacy, security and records owners before extracting anything. From Customer Abuse Is a WHS Signal, Not Just a Conduct Problem.

No. An AI safety system is a control measure, not a transfer of duty. Safe Work Australia is explicit that a PCBU cannot contract out of their responsibility and a duty cannot be transferred to another person. The vendor has its own upstream duty over the product, but that sits alongside yours. Verifying the system works on your site is you discharging your duty. From An AI Safety Control Is Still Just a Control. The Duty to Verify It Stays With You.

No. Notifying the regulator of a notifiable incident is a separate statutory duty, and it is a decision and a phone call humans make first. The skill file carries an explicit guardrail that drafting a communication never substitutes for, or delays, regulator notification. From The Safety Comms Skill File: Alerts Workers Actually Read

No. A quiz can provide evidence of knowledge or recall. Competence may require practical demonstration, observation, formal assessment, licence evidence or supervised performance appropriate to the task. A competent person selects the method and decides whether the evidence is sufficient. From AI Can Draft WHS Training. It Cannot Verify Competence.

No. Safe disclosure is a local workflow label, not a statutory category, a confidentiality promise or immunity from human-led investigation. Suspected serious misconduct, security events and privacy incidents still go to authorised people under existing processes, with the facts assessed by humans. The point of the firewall is the quality of the risk picture, not a guarantee that conduct will never be examined. From Shadow AI Is Already Changing the Work. Govern the Reality, Not the Approval Register.

No. Victoria has not implemented the model WHS laws. Under section 144 of the Occupational Health and Safety Act 2004 (Vic), an officer may be liable where the body corporate's contravention is attributable to the officer's failure to take reasonable care, with regard to matters including what the officer knew and the extent of their ability to make or participate in relevant decisions. The six-part file can still be useful governance practice in Victoria, but it is not a Victorian statutory checklist. From Your AI Dashboard Is Not Due Diligence: What Officers Need to Know, Ask and Record

Not uniformly. The OAIC explains that public and private sector employee records are treated differently and that the private sector employee records exemption has defined limits. Where the Australian Privacy Principles apply, personal-information inputs and outputs attract privacy obligations, and the OAIC recommends against entering personal, particularly sensitive, information into public generative AI tools. From The First Manager Message After an Incident Can Help or Harm. Govern the Draft.

No. Victoria has a separate OHS regime. For psychosocial controls, WorkSafe Victoria's compliance code explains the distinct regulation 16 triggers, including relevant alterations, new or additional hazard information, a report of a psychological injury or psychosocial hazard, a notifiable incident involving a psychosocial hazard, inadequate control and an HSR request. Do not copy Commonwealth section labels into a Victorian register. From A Risk Register Is Not Current Because AI Reworded It.

No. It is an occurrence gate, not an effectiveness verdict. Training may happen without building competence, an escalation screen may be live but poorly designed, and a consultation meeting may occur without satisfying applicable requirements. Those are later questions answered with different evidence. From An AI-Written WHS Record Is Not Audit Evidence Until You Prove the Work Happened.

Yes. Section 19 of the WHS Act 2011 (NSW) defines health to include psychological health and requires safe systems of work. An AI tool that allocates, paces or monitors work is a system of work, so the duty applies to the psychosocial risks it creates, the same as any other aspect of work design. From AI Work Allocation and Psychosocial Risk: The WHS Duty NSW Teams Already Hold

Not in model-law terms. Victoria has an OHS regime rather than the model WHS Act. Section 21(3) of the Occupational Health and Safety Act 2004 (Vic) extends the employer duty to independent contractors and their employees for matters over which the employer has control, and section 35 extends consultation within the same boundary. Map the Victorian framework rather than copying model-law section labels. From The Contractor Used AI to Write the Safety Pack. Your WHS Duty Did Not Move.

It must never be allowed to. Under section 38 of the model WHS Act, notice of a notifiable incident goes to the regulator immediately after becoming aware, in the fastest way possible, which can be by telephone, email or an online form where available. The skill file in this article prints an urgency line at the top of every output: if this may be notifiable, call first, the document waits. From An Incident Triage Skill File That Never Decides Notifiability

No. NSW Parliament passed an amendment introducing an express digital work systems duty, but its operative provisions commence on a day to be appointed by proclamation, which has not yet occurred. Until then, manage AI-related psychosocial risk under the existing duties in the WHS Act and the WHS Regulation, which already apply. From AI Work Allocation and Psychosocial Risk: The WHS Duty NSW Teams Already Hold

AI reads across many de-identified incident records and proposes patterns a human would take hours to assemble. Three supported uses hold up: de-identified trend and leading-indicator analysis, ICAM-style causal support for a single incident, and drafting investigation summaries with blanks. The output is a hypothesis, never a finding, and a competent person tests and decides. From AI for Incident Analysis and Leading Indicators: A Human-in-the-Loop WHS Playbook

Set the boundary in writing in your project instructions and prompts so every rating field returns blank, marked competent person to determine. If the drafting tool ever returns a populated risk-level column, treat that as a prompt defect, delete it, and put the rating back in human hands. The competent person rates and signs. From AI-Assisted Psychosocial Risk Assessment: A WHS Governance Workflow That Keeps the Sign-Off Human

Test it against your own site conditions, not the vendor demo footage, and probe two failures in opposite directions. The false negative is the dangerous miss, the worker in the zone the system did not flag. The false positive is the nuisance alert that trains everyone to ignore it. Vendor accuracy figures do not close this out. Your own test on your own site does. From An AI Safety Control Is Still Just a Control. The Duty to Verify It Stays With You.

Build challenge families around each ledger row rather than a generic answer-quality test set. Change spelling, tone, location, job role, time of day and the amount of personal information. Join an authorised question to a prohibited one, and ask the same question after the source expires or the route is unavailable. The failure you are hunting is an authority leak from approved retrieval into judgement. From If Your Safety Chatbot Can Answer Freely, It Can Answer Wrongly.

With workers who will actually use the information. WorkSafe Victoria's code recommends asking people to restate the information in their own words, identify equipment or sign meanings, answer content questions or demonstrate the task, and says questions should be phrased so the employee gives more than a yes or no answer. Record results by language and channel, not by worker performance. From AI Translation Can Multiply a Safety Error Across Every Language.

AI proposes candidate contributing factors at the four ICAM levels for a single de-identified incident: absent or failed defences, individual or team actions, task or environmental conditions, and organisational factors. The investigator then tests each factor against the evidence, discards weak ones, and adds what the model missed. The model widens the search; the competent investigator decides what is true. From AI for Incident Analysis and Leading Indicators: A Human-in-the-Loop WHS Playbook

Victoria applies the Occupational Health and Safety Act 2004 instead of the model WHS Act, using an employer and employee regime. Section 72 requires a committee within three months after an HSR request or if regulations require, with no model-law request trigger for five workers. At least half the members must be employees and, so far as practicable, HSRs or deputies. WorkSafe Victoria says responsibility for health and safety issues rests with the employer. From Your WHS Committee Pack Should Ask for Decisions, Not Bury Them in Updates.

Victoria has its own Occupational Health and Safety (Psychological Health) Regulations 2025, in force since 1 December 2025. Regulation 15 requires employers to eliminate psychosocial risk so far as reasonably practicable, then reduce it using work-management, plant, systems-of-work, work-design or environmental measures. Information, instruction or training may only be used on its own where none of those measures is reasonably practicable, and in a combination it must not be the predominant control. Do not paste Commonwealth regulation numbers into a Victorian assessment. From AI Made the Queue Faster. It Also Made the Job Harder.

Measure the quality of the exercise and action closure, not the number of scenarios generated. Useful measures include new dependencies identified, participation across shifts, actions supported by evidence, time to close actions and successful verification at the next test. From AI Can Stress-Test an Emergency Plan. It Cannot Run the Drill.

Never paste real personal, claim, health or incident data into a model that is not an approved enterprise instance, and public consumer chatbots do not qualify. Before any export leaves your safety system, strip names, employee IDs, claim numbers, exact dates, and re-identifying free text, replacing them with placeholder tokens such as [EMPLOYEE_NAME], [CLAIM_NUMBER] and [SITE]. From AI for Incident Analysis and Leading Indicators: A Human-in-the-Loop WHS Playbook

No. Safe Work Australia's PCBU information sheet on working from home says homes and living arrangements vary, so risk controls will vary, and recommends a combination of methods: surveys, tools and checklists, prompt reporting, consideration of tasks and resources, monitoring the work environment and reviewing incident, complaint and consultation information. It also notes a health and safety professional may be needed where potential risks warrant specialist assessment. From A Green Home-Office Score Does Not Clear the Risk.

No. The duty is to identify hazards, control risks so far as is reasonably practicable, maintain controls and review them in specified circumstances. Safe Work Australia's model Code says a PCBU may prepare a risk register, and the Commonwealth psychosocial Code says the recording method should suit the organisation's circumstances, while warning that without a written record the duty holder would need to demonstrate by other means how duties were met. From A Risk Register Is Not Current Because AI Reworded It.

No. Australian WHS law does not create a standalone hazard called AI pace. Safe Work Australia's guidance says AI and digital technologies can introduce or contribute to physical and psychosocial hazards, so the question is whether the technology changes recognised exposures such as high job demands, low job control, poor work design or inadequate support, alone or in combination. From AI Made the Queue Faster. It Also Made the Job Harder.

No. An HSR is an elected worker representative with an ongoing role for their work group and no suspected-breach threshold for asking about hazards and risks. A WHS entry permit holder is a union official whose inquiry entry power depends on a reasonably suspected contravention affecting relevant workers, with statutory conditions on inspecting work systems and directly relevant documents at the workplace. From An HSR Asked How the AI Tool Changes Work. A Vendor Slide Is Not an Answer.

No. It is a diagnostic only. A second system can repeat the same error, and a literal return can distort a sound contextual choice. Review critical meaning in a separate instruction ledger instead, comparing who must act, what is required or prohibited, when it applies and in what order, quantities and contacts, the hazard addressed, and the route when the instruction cannot be followed. From AI Translation Can Multiply a Safety Error Across Every Language.

No. There is no regulation 55E called cumulative AI change load. Overlapping changes instead increase the duration, frequency and interaction of recognised psychosocial hazards such as poor organisational change management, high job demands, low role clarity, low job control and poor support. The portfolio view is an evidence method for applying regulations 55A to 55D to the work as experienced. From One AI Rollout Is Change. Ten Overlapping Updates Are an Exposure Pattern.

No. Comcare describes section 27 as a proactive officer duty that is separate from the organisation's own duties, and Safe Work Australia's officer guideline says the officer must take an active and inquisitive role. A dashboard is one input. The officer still needs to trace favourable claims to source records, test them against operational reality and worker experience, and record the decision that followed. From Your AI Dashboard Is Not Due Diligence: What Officers Need to Know, Ask and Record

Yes. Introducing a system that changes how work is allocated, paced, measured, monitored, or decided is a change to the work. The primary duty of care and the duty to consult workers, which include psychological health, already cover it. NSW has now made the AI case express in statute, and the model WHS jurisdictions, Victoria, and the Comcare scheme reach it through existing duties. From Rolling Out AI Is a Workplace Change: Consult and Risk-Assess First

No, and conflating them is dangerous. Screening past records for patterns is not live threat detection, an emergency response, an incident investigation or a notifiability decision. Ordinary reporting and emergency routes stay open, and a person receiving an immediate threat follows the approved human escalation and emergency procedure. From Customer Abuse Is a WHS Signal, Not Just a Conduct Problem.

An approved code under section 274 of the WHS Act is not a standalone offence, but it is admissible in court as evidence of known hazards and reasonably practicable controls. Following the Code is the most defensible position. Departing from it shifts the burden to you to prove your alternative approach was at least as effective. From AI Work Allocation and Psychosocial Risk: The WHS Duty NSW Teams Already Hold

No. Safe Work Australia's July 2026 guidance says AI and digital technologies can introduce or contribute to physical and psychosocial hazards, and that existing risk-management duties apply, in consultation with workers and their representatives. There is no statutory AI shortcut, and pushed software updates can change WHS risks after deployment. From Your AI Dashboard Is Not Due Diligence: What Officers Need to Know, Ask and Record

No. Automatic rejection is as wrong as automatic approval. AI provenance changes the reliability inquiry, not the duty allocation. Ask what the tool did, which sources it used and which competent person checked each material output against current sources and the work as performed, then test the controls at the shared-duty interfaces. From The Contractor Used AI to Write the Safety Pack. Your WHS Duty Did Not Move.

Keep both. Rates against a suitable contact-volume denominator stop the busiest team looking most exposed when a smaller channel has a higher concentration of abuse. Raw totals still matter because a high rate and a high total load answer different control questions. From Customer Abuse Is a WHS Signal, Not Just a Conduct Problem.

Queue velocity (how quickly the next item arrives and whether a worker can pause the flow), sustained cognitive effort (how much remaining work needs ambiguity resolution or high-consequence judgement), task switching (how often the worker changes customer, system or decision frame), review burden (what must be checked before relying on AI output, and whether that time is recognised), and recovery time (whether planned breaks and resets survive the faster queue). Keep the signals separate; one favourable result can hide a harmful one. From AI Made the Queue Faster. It Also Made the Job Harder.

Reasonable steps to acquire and keep current WHS knowledge; to understand the operations and their hazards and risks; to ensure appropriate resources and processes are available and used; to ensure information about incidents, hazards and risks is received, considered and responded to in a timely way; to ensure processes for complying with duties are in place and implemented; and to verify that those resources and processes are provided and used. Comcare notes the list is non-exhaustive. From Your AI Dashboard Is Not Due Diligence: What Officers Need to Know, Ask and Record

A pattern lane that collects work group, task, tool category, information class, frequency, decision influence and work effects at group level. An urgent-event lane that routes suspected exposure of personal or confidential information, unsafe automated action or decision substitution straight to the authorised WHS, privacy, security or legal process. And a separate human-authorised conduct lane with its own purpose, access and procedural safeguards. From Shadow AI Is Already Changing the Work. Govern the Reality, Not the Approval Register.

A source record, meaning a controlled, identifiable input created by or close to the activity, with owner, date, version, location and limits. An accountable human confirmation from a named person with direct knowledge who confirms the underlying activity rather than approving the wording. And an operational trace or observation, a separate signal showing the activity entered the work system. Do not generate a second document from the same prompt and call it corroboration. From An AI-Written WHS Record Is Not Audit Evidence Until You Prove the Work Happened.

Four in particular. A rewrite can make an old assumption sound present, merge distinct controls into one vague statement, fill an evidence gap with plausible prose and erase the reason a human previously accepted or rejected a change. None of those acts tests whether the control operates in the work. From A Risk Register Is Not Current Because AI Reworded It.

Assemble the decision packet from approved, de-identified records, match figures to controlled sources, expose missing cut-off dates and red-team whether the ask is clear. It must not decide the control, speak for workers, resolve disagreement or declare consultation complete, and personal or medical information stays out under section 79 unless the worker consents or the material is de-identified. From Your WHS Committee Pack Should Ask for Decisions, Not Bury Them in Updates.

AS/NZS ISO 45001:2018 replaced the National Audit Tool as the prevention-management-system standard from that date. Applicants must provide evidence of ISO45001 certification or an audit report demonstrating alignment that meets the Commission's requirements, and licensees must demonstrate annually that they maintain a certified or aligned system. The SRCC's Licence Compliance and Performance Model says the transition began on 1 July 2026 and is expected to finish by 30 September 2028. From An AI-Written WHS Record Is Not Audit Evidence Until You Prove the Work Happened.

The 2025 amendments extend notification to violent incidents that expose a person to a serious risk of psychological harm, work-related suicides and attempted suicides, and extended worker absences of 15 or more consecutive calendar days due to a work-related injury or illness, and add an express duty to preserve evidence alongside the incident site. Safe Work Australia notes these requirements apply only as each jurisdiction adopts them, so check with your WHS regulator. From An Incident Triage Skill File That Never Decides Notifiability

Strip all personal, claim, health, and incident data to placeholder tokens such as EMPLOYEE_NAME, CLAIM_NUMBER, INCIDENT_ID, TEAM, ROLE, SITE, and DATE before anything reaches the model. Aggregate into themes and counts where possible. De-identification is a control you apply at source, not a step the model performs for you. From AI-Assisted Psychosocial Risk Assessment: A WHS Governance Workflow That Keeps the Sign-Off Human

Exclude real employee, incident, client, medical and witness information unless the tool and data pathway are specifically approved. Use placeholders such as [EMPLOYEE_NAME], [TEAM], [SITE], [TASK], [HAZARD], [CONTROL] and [INCIDENT_ID], and de-identify before upload, not after. From AI Can Draft WHS Training. It Cannot Verify Competence.

Do not paste anything that identifies people, the client, or commercially sensitive site detail into a public AI tool while drafting. A SWMS is about the work and the controls, and it does not need names to do its job. Strip identifiers to placeholder tokens before anything reaches the model. From AI Can Draft a SWMS in Seconds. The Site Walk and the Consultation Cannot Be Automated.

That it remains effective, including by being fit for purpose, suitable for the nature and duration of the work, and installed, set up and used correctly. A register row that has not been tested against operation cannot show any of those three limbs, no matter how current its wording looks. From A Risk Register Is Not Current Because AI Reworded It.

For the Commonwealth jurisdiction, regulation 39 of the Work Health and Safety Regulations 2011 requires information, training and instruction provided to a worker to be suitable and adequate having regard to the nature of the work, the risks associated with the work at the time, and the control measures implemented. So far as is reasonably practicable, it must also be provided in a way that is readily understandable. Check the equivalent provision in your own jurisdiction. From AI Can Draft WHS Training. It Cannot Verify Competence.

In determining which control measures to implement, a PCBU must have regard to all relevant matters. Those include the duration, frequency and severity of exposure, how hazards may interact or combine, job demands, systems of work including how work is managed and supported, and the information, training, instruction and supervision provided. A portfolio map supplies that evidence at the work-group level. From One AI Rollout Is Change. Ten Overlapping Updates Are an Exposure Pattern.

Its AI guidance lists, as an example risk, automation of routine or basic tasks resulting in workers having a greater proportion of tasks which require extensive thought and focus, and labels that result work intensification. Its suggested controls include designing automation so workers can alter the pace of work, change tasks or pause work to take breaks as needed. From AI Made the Queue Faster. It Also Made the Job Harder.

It was assented to on 18 February 2026, but as at 31 July 2026 most operative provisions had not commenced. The future power to require reasonable assistance to access and inspect a digital work system is an entry permit holder power, not an HSR power, and SafeWork NSW says the remaining provisions await proclamation after the Digital Work Systems Guidelines are published. From An HSR Asked How the AI Tool Changes Work. A Vendor Slide Is Not an Answer.

The Work Health and Safety Amendment (Digital Work Systems) Act 2026, assented on 18 February 2026, defines a digital work system as an algorithm, artificial intelligence, automation, or online platform. It extends the primary duty and adds a new section 21A requiring a PCBU to consider whether work allocation creates excessive workloads, unreasonable performance metrics, excessive monitoring or surveillance, or discriminatory decisions. From Rolling Out AI Is a Workplace Change: Consult and Risk-Assess First

It saves time and adds consistency on the mechanical work: synthesising free-text, mapping to categories, and producing a structured first draft. The auditable trail from de-identified data to a drafted assessment with blank ratings is a governance asset. It does not shortcut the duty, which still rests on the organisation and competent people. From AI-Assisted Psychosocial Risk Assessment: A WHS Governance Workflow That Keeps the Sign-Off Human

Section 14 of the Work Health and Safety Act 2011 provides that a duty cannot be transferred. Sections 47 to 49 require consultation, so far as is reasonably practicable, with workers directly affected by a work health or safety matter. Where the chatbot provides required information or instruction, regulation 39 requires it to be suitable and adequate having regard to the nature of the work, the nature of the risks and the control measures implemented. From If Your Safety Chatbot Can Answer Freely, It Can Answer Wrongly.

The source identifier, locale and script, method and tool version, glossary, reviewer, exceptions, approval, channel, comprehension evidence and withdrawal trigger. It sits beside the source lock, which records the source title, owner, jurisdiction, version and immutable identifier, the intended work group and conditions of use, the critical instructions and prohibitions, and the expiry and change triggers. Together they are the chain of custody for meaning. From AI Translation Can Multiply a Safety Error Across Every Language.

As at 31 July 2026, SafeWork NSW says only the provisions supporting guideline development have commenced. All other provisions, including the new work-allocation duty and the entry-permit-holder assistance power itself, commence on proclamation, and that start date must be at least one month after the guidelines are published. Treat those as future obligations rather than current law. Existing WHS duties still apply. From Shadow AI Is Already Changing the Work. Govern the Reality, Not the Approval Register.

Personal or medical information without the worker's consent, unless it neither identifies the worker nor could reasonably be expected to lead to identification, and material protected by legal professional privilege. Commercial sensitivity is different: a vendor confidentiality clause is not an express statutory exception, so give the HSR the risk-relevant substance through an extract, briefing or controlled inspection and record the precise basis for anything withheld. From An HSR Asked How the AI Tool Changes Work. A Vendor Slide Is Not an Answer.

Keep out names, contact details, health or disability information, security credentials, detailed access data and sensitive floor plans unless an approved environment and clear authority exist. Use role and site placeholders such as [SITE] and [WARDEN_ROLE] wherever possible, and de-identify before anything enters the model. From AI Can Stress-Test an Emergency Plan. It Cannot Run the Drill.

A documented sample of green results selected for human review independently of worker performance, compared with equipment-delivery status, unresolved hazard reports and consultation feedback. If reviewers find missed hazards, the questions and routing rules are revised. It tests the screening method, not the worker. From A Green Home-Office Score Does Not Clear the Risk.

A leading indicator is a signal that predicts harm before it happens, such as a rising rate of near-misses of a particular type, rather than a lagging indicator like a lost-time injury that records harm after the fact. AI can surface candidate leading indicators across de-identified data for a human analyst to confirm against further evidence. From AI for Incident Analysis and Leading Indicators: A Human-in-the-Loop WHS Playbook

Under section 35 of the model WHS Act, a notifiable incident arising from the conduct of a business or undertaking is the death of a person, a serious injury or illness, a dangerous incident, or a violent incident. The 2025 amendments also added notifiable extended absences of 15 or more consecutive calendar days due to a work-related injury or illness, and notifiable work-related suicides and attempted suicides. Adoption varies by jurisdiction, so check with your WHS regulator. From An Incident Triage Skill File That Never Decides Notifiability

A reusable markdown instruction file that encodes a team standard for safety alerts, toolbox talk intros and bulletins in six sections: purpose, when to use, inputs required, method, output format and guardrails. Loaded into an AI workspace, it makes every draft start at the standard: plain language, site-specific facts, one clear action, no blame. From The Safety Comms Skill File: Alerts Workers Actually Read

It is the record that defines what the chatbot may say. For every permitted response it captures the intent identifier, permitted response form, approved source and version, exact passage, excluded questions, jurisdiction, owner, expiry, change trigger and route. If the ledger does not authorise an answer, the chatbot uses a fixed escalation message instead of improvising. From If Your Safety Chatbot Can Answer Freely, It Can Answer Wrongly.

A short record that sits between the contract register and the risk register. Each row covers one matter where the organisations' work, people, premises or controls touch, and captures the duty-holder basis, influence and control, evidence lineage including AI involvement, the control hand-off, worker consultation, verification evidence and the change trigger that reopens the row. From The Contractor Used AI to Write the Safety Pack. Your WHS Duty Did Not Move.

Victoria requires its own analysis. The Occupational Health and Safety (Psychological Health) Regulations 2025 have been in force since 1 December 2025, and regulations 14 to 16 require employers to identify hazards, control risks and review controls. Under section 35 of the Occupational Health and Safety Act 2004, consultation applies to directly affected or likely directly affected employees, involving the HSR where employees are represented. Use Victorian employer and employee terminology, not Commonwealth provisions. From One AI Rollout Is Change. Ten Overlapping Updates Are an Exposure Pattern.

Section 77 gives a health and safety committee functions that facilitate co-operation between the PCBU and workers in instigating, developing and carrying out health and safety measures. Section 68 gives a health and safety representative separate functions for a work group: representing workers, monitoring measures, investigating complaints and inquiring into apparent risks. An HSR may sit on the committee, but ordinary committee members do not acquire HSR powers. From Your WHS Committee Pack Should Ask for Decisions, Not Bury Them in Updates.

The channel, journey stage or time window, never the individual. A channel-level pattern points at work design, which is where the control sits. A person-level score points at someone to manage or sanction, which raises separate legal, privacy, conduct and fairness questions and does nothing to reduce the exposure. From Customer Abuse Is a WHS Signal, Not Just a Conduct Problem.

An owner, a test, an expiry and a workable alternative. The control card should record the activity, the immediate boundary, the safe alternative, any temporary workload adjustment, the human owner, the consultation step, the evidence needed and the review trigger. Stopping a time-saving workaround while leaving the same queue, staffing and deadline in place simply intensifies the condition that produced it. From Shadow AI Is Already Changing the Work. Govern the Reality, Not the Approval Register.

Diagnose an injury, infer disability, reject a flexible-work arrangement, direct office attendance, rank workers by supposed vulnerability or clear the arrangement. A report of discomfort may need a conversation, an ergonomic assessment or another authorised pathway, and those calls belong to appropriately skilled people, not a scoring method. From A Green Home-Office Score Does Not Clear the Risk.

The model can watch, detect and flag tirelessly. It cannot decide what to control, choose where the control sits in the hierarchy, verify it works in your conditions, weigh the surveillance it creates against the harm it prevents, respond to what it flags, or carry the duty when it misses. Those are judgements the law puts on a person and an organisation, not on a sensor. From An AI Safety Control Is Still Just a Control. The Duty to Verify It Stays With You.

AI is genuinely useful for the blank page: laying out the required format, listing standard hazards for a task type so you are less likely to miss one, surfacing candidate controls, and rewriting dense safety language into plain English, which is what regulation 299(3)(b) asks for. Used as a drafting assistant against your own template and site knowledge, it removes the slow part and leaves the judgement to you. From AI Can Draft a SWMS in Seconds. The Site Walk and the Consultation Cannot Be Automated.

Workers may type names, health details or allegations even when the bot did not ask. Use an approved environment, warn workers before entry not to provide unnecessary personal or sensitive information, minimise log contents, separate safety escalation records from analytics, define access and retention, and give a private human channel. The OAIC advises organisations not to enter personal information, particularly sensitive information, into publicly available generative AI tools. From If Your Safety Chatbot Can Answer Freely, It Can Answer Wrongly.

The system's purpose and actual use, a plain-language workflow with override and escalation points, the hazard and risk assessment including psychosocial hazards, the controls and evidence they operate, testing results and known limitations, de-identified complaint, incident, override and workload trends, training and support arrangements, consultation records, and the change log including vendor updates. From An HSR Asked How the AI Tool Changes Work. A Vendor Slide Is Not an Answer.

Documented override triggers: reported pain or discomfort, a relevant prior injury or individual need, missing or ambiguous answers, a move or renovation, changed equipment or tasks, an incident, or a worker or HSR concern. Free-text comments must be read, not ignored because the scored fields passed. Every override routes to a named practitioner, assessor, manager or specialist. From A Green Home-Office Score Does Not Clear the Risk.

Preserve the contradiction; it is evidence, not an editing problem. A central log may say a protocol reached every team while sampled workers reveal an overnight team lacked access. The contradiction changes the claim's scope. Record it as corroborated with a scope limit rather than flattening it into a clean national implementation statement. From An AI-Written WHS Record Is Not Audit Evidence Until You Prove the Work Happened.

A role-coded decision line with five lanes: RECOMMEND for the committee's collective position, AGREE COMMITTEE PROCEDURE for internal meeting rules, RECORD WORKER OR HSR VIEW where consultation remains open, AUTHORISED MANAGER DECISION for the person or forum with internal authority to decide and resource action, and SEPARATE HSR PATHWAY for statutory steps that remain with the HSR. From Your WHS Committee Pack Should Ask for Decisions, Not Bury Them in Updates.

Five lines in order: the verified fact stated only to the level authorised for the audience, the immediate work setting change, the approved support offer, a statement of choice and privacy, and the next human contact with owner, channel and time. It should not diagnose, speculate, demand disclosure or promise an unapproved outcome. From The First Manager Message After an Incident Can Help or Harm. Govern the Draft.

The site walk, the consultation, the judgement that a control is adequate, the decision to proceed or stop work under regulation 300, and the signature of the person who carries the duty. AI cannot hold the primary duty of care. That responsibility rests with the person conducting a business or undertaking and its officers, and it does not move to a tool because the tool produced the first draft. From AI Can Draft a SWMS in Seconds. The Site Walk and the Consultation Cannot Be Automated.

When the control is not effective; before a workplace change likely to give rise to a new or different risk the measure may not effectively control; when a new relevant hazard or risk is identified; when consultation indicates a review is necessary; and when a health and safety representative makes a qualifying request. Safe Work Australia's AI guidance names a software update as an example of a change that can trigger review. From A Risk Register Is Not Current Because AI Reworded It.

When material facts are disputed, the audience is unclear, the message could identify an affected person, an emergency response is active or the approved escalation owner has not reviewed it. Speed is not the goal; safe, useful contact is, and message drafting must never delay emergency action or the separate human-led assessment of notification duties. From The First Manager Message After an Incident Can Help or Harm. Govern the Draft.

Almost every detect-and-alert AI safety product is an administrative control. It does not remove, isolate or engineer out the hazard, it watches and relies on a human noticing the alert and acting in time. Safe Work Australia ranks administrative controls and PPE as the least effective because they rely on human behaviour and supervision, so the tool sits near the bottom where reliability is lowest. From An AI Safety Control Is Still Just a Control. The Duty to Verify It Stays With You.

The same markdown file runs as project instructions with knowledge files on ChatGPT and Claude plans that include projects, installs natively where a plan supports the Skills features both vendors now ship, and adapts to Microsoft 365 Copilot by pasting the contents into an agent Instructions field in Agent Builder. From The Safety Comms Skill File: Alerts Workers Actually Read

The Safe Work Australia model Code of Practice (July 2022) is the practical method, sitting over model WHS Regulations 55A to 55D and the section 19 primary duty in the model WHS Act. Adoption varies: NSW took effect 28 May 2021, Comcare covers Commonwealth employers, and Victoria and Western Australia run distinct arrangements. From AI-Assisted Psychosocial Risk Assessment: A WHS Governance Workflow That Keeps the Sign-Off Human

Use a stop class for immediate danger or emergencies, medical or fitness matters, notifiability or site preservation, cease-work or HSR rights, individual risk assessments, control adequacy, legal interpretation, identifiable allegations, stale or conflicting sources, and safety-critical instructions outside current approved text. The stop class is a designed safety response. It identifies immediate contacts without pretending the chatbot assessed urgency, entitlement or legal status. From If Your Safety Chatbot Can Answer Freely, It Can Answer Wrongly.

Clauses 55A to 55D of the WHS Regulation 2025 (NSW) define psychosocial hazards and risks and require PCBUs to manage them using the hierarchy of controls in clause 36. Clause 55D lists the matters to weigh, including exposure duration and severity, how hazards combine, and the design and systems of work. From AI Work Allocation and Psychosocial Risk: The WHS Duty NSW Teams Already Hold

The definition includes directors and secretaries, people who make or participate in decisions affecting the whole or a substantial part of the business, and people who can significantly affect its financial standing. A senior title does not settle it, and a WHS manager is not automatically an officer merely because they advise on safety. Status turns on the statutory definition and the actual role. From Your AI Dashboard Is Not Due Diligence: What Officers Need to Know, Ask and Record

The review group should include the competent WHS lead and the people who understand the work, including workers, health and safety representatives, wardens, facilities, shared-site duty holders and emergency services where appropriate. Consultation with workers and HSRs when making and reviewing emergency plans is part of the duty. From AI Can Stress-Test an Emergency Plan. It Cannot Run the Drill.

Because the first hour of an incident is the worst time to be writing instructions. A skill file captures the structure, the guardrails, and the never-decide rule once, is tested against fictional scenarios in calm conditions, and then runs the same way every time under pressure, on whichever platform your team uses. From An Incident Triage Skill File That Never Decides Notifiability

They open with filler, bury the learning, hide the action and carry a flavour of blame, so workers skim and change nothing. Site-specific facts, one clear action and plain language are what get read, and that standard is exactly what the skill file encodes and enforces on every draft. From The Safety Comms Skill File: Alerts Workers Actually Read

Regulation 299(3) requires a SWMS to be prepared taking into account circumstances at the workplace and to be readily understandable to the people who use it. A document built from a task description and a training corpus has taken account of no workplace at all. SafeWork NSW states a SWMS must be site-specific, so a generic AI draft is non-compliant by construction until reviewed and amended for the actual site. From AI Can Draft a SWMS in Seconds. The Site Walk and the Consultation Cannot Be Automated.

Because the harms from a badly introduced AI system are mostly psychological, and psychological health is inside the duty. The four risks NSW names map onto recognised psychosocial hazards: workload onto job demands, metrics onto job control, surveillance onto intrusive surveillance, and discriminatory decisions onto organisational justice. Poor change management is itself a recognised hazard. From Rolling Out AI Is a Workplace Change: Consult and Risk-Assess First

Because the change is to the work, not just the software list. Undisclosed use can alter task order, throughput expectations, cognitive load, access to supervision, error pathways, role clarity and the point at which a human sees source material. Safe Work Australia's managing-risks guidance names increased pace, higher workload, cognitively demanding residual work, low job control, poor support, lack of role clarity and poor organisational change management among its examples. From Shadow AI Is Already Changing the Work. Govern the Reality, Not the Approval Register.

To test false negatives. A validation set of flagged records alone only tells you about the model's positives. The no-signal sample is where coded threats, sarcasm and language the model handles poorly show up, and those are exactly the exposures workers report and systems miss. From Customer Abuse Is a WHS Signal, Not Just a Conduct Problem.