Executive Summary
The Challenge: AI tools for generating safety & risk management outputs such as Safe Work Method Statements are arriving faster than industry understanding. Before these technologies become standard practice, we need to ask better questions about responsible adoption.
The Approach: This workshop brought together diverse stakeholders to explore AI's role in the SWMS risk assessment process through human-centered design methodology.
What We Discovered
- The industry is at the beginning of its AI maturity journey and fundamental questions remain unanswered
- Given the difference of opinion when asked, interestingly the intent of SWMS is contested so before we automate, we need clarity on purpose
- Different stakeholders have conflicting needs
- This isn't specific to SWMS—every safety process faces the same questions about AI integration
The Central Finding
We convened to discuss AI in SWMS generation. What we discovered is that we first need deeper conversations about SWMS themselves—and indeed, about all our safety processes. This clearly, is not a technology problem; it's a safety and work practice problem that technology might amplify or address, depending on how thoughtfully we proceed.
The Path Forward: This report documents a snapshot of where industry likely is today—early, exploratory, asking foundational questions. The opportunity shouldn't be rushing to adopt AI tools. It should be building community consensus on responsible innovation principles that protect what matters most: keeping people safe.
Why This Workshop, Why Now?
Safe Work Month 2025: Assess Risks
This workshop aligned with 2025 Safe Work Month's focus on risk assessment. The timing was deliberate—as AI tools proliferate across industries, safety professionals and the businesses they serve need to examine how these technologies intersect with established risk management practices.
SWMS were chosen as a case example because they represent a key safety process that's simultaneously:
- Legally mandated for high-risk construction work
- Widely used across multiple industries (especially if we use them as a proxy for all field risk assessment)
- Currently problematic—often experienced as "tick-box" compliance rather than genuine engagement process
- Targeted by AI developers as an obvious automation opportunity
Beyond SWMS
While we focused on SWMS, the questions we explored apply to ANY safety process: incident investigations, job safety analyses, risk assessments, permit systems, audits, inspections, toolbox talks. If we can't articulate the genuine purpose of these activities and how different stakeholders experience them, we can't responsibly deploy AI to support them.
"AI capabilities are real, they are here to stay, and we need to move the discussion to thinking more about what this means and what we want it to mean."
— Ethan Mollick
This workshop operated from that premise. We didn't debate whether AI works (with the right solution design it can work very well). We asked harder questions about when AI applications are uplifting versus detrimental in safety-critical contexts.
Returning to First Principles: What is the Intent of a SWMS?
Before exploring AI applications, we grounded the discussion in a fundamental question: What are SWMS actually supposed to achieve?
The Regulatory Intent
According to WHS Regulations, a SWMS is:
- A document identifying high-risk construction work, associated hazards, and control measures
- An administrative control supporting higher-order risk elimination/minimisation
- A tool to help PCBUs and workers confirm and monitor controls (not a procedure itself)
- Required to be site-specific, consulted on, and kept accessible
- Must be prepared before work starts and followed during execution
The Reality Gap
Workshop participants acknowledged tension between what regulations require and what actually happens in practice:
The Ideal
- Systematic risk identification
- Worker consultation and understanding
- Site-specific hazard capture
- Living document that adapts
- Tool that enables safe work
The Reality
- Copy-paste from templates
- Minimal consultation
- Generic, not site-specific
- Sign and file
- "Tick-box" compliance exercise
Workshop North Star
Participants agreed that before evaluating any AI tool, we must judge it against the genuine intent of the process - in this case a SWMS: Does this help workers understand and control risks, or does it just make compliance paperwork faster?
Speed without substance isn't innovation—it's automation of a broken system.
Understanding Stakeholder Experiences: The Persona Lens
We explored SWMS experiences through different stakeholder lenses because AI implementation choices will impact each group differently. What solves one persona's problem may create new problems for another.
Personas Developed
Groups created representations of key stakeholder experiences:
- Frontline workers (including apprentices, ESL workers)
- Supervisors and leading hands
- Small subcontractors (2-5 employees)
- Principal contractors managing multiple SWMS
- Corporate safety managers and advisors
- Regulators and inspectors
Cross-Cutting Themes
Universal Pain Points:
- Time pressure — "Need it done yesterday"
- Complexity vs. usefulness — Long documents nobody reads
- Disconnect from reality — Doesn't reflect actual site conditions
- Consultation theatre — Going through motions rather than genuine engagement
- Accountability confusion — Who's responsible when it's inadequate?
Conflicting Needs:
- Workers need simplicity and clarity → Regulators need comprehensive documentation
- Contractors need speed and efficiency → Safety requires thoroughness and site-specificity
- Principals need standardisation across subcontractors → Work reality requires flexibility and adaptation
The "Tick-Box" Reality
Participants were honest: SWMS have become tick-box compliance documents rather than safety tools for many organisations. This isn't because people don't care—it's because of competing pressures, time constraints, unclear value propositions, and systemic issues that predate any AI discussion.
Critical question: Will AI tools address these foundational issues, or will they enable us to simply produce tick-box documents faster?
The AI Technology Landscape
We introduced three emerging AI applications to ground the discussion in concrete use cases:
1. AI SWMS Generators
What they do: Generate complete SWMS documents from basic inputs (work type, location, equipment, etc.)
Promise: Speed, consistency, comprehensive hazard identification, template standardisation
Current state: Multiple vendors developing tools; some already in market
2. AI Quality Assessors
What they do: Review SWMS for gaps, compliance with regulations, proper control hierarchy, completeness
Promise: Consistent quality checking, learning support for inexperienced users, regulatory alignment
Current state: Emerging technology; primarily in pilot/development phase
3. Voice AI SWMS Coach
What they do: Convert natural conversations (toolbox talks, planning discussions) into structured SWMS format
Promise: Capture authentic worker knowledge, preserve consultation process, reduce administrative burden
Current state: Conceptual; voice AI capabilities exist but application to SWMS is mostly nascent
What the Workshop Revealed
Participants struggled to quickly assess these technologies. Not because the concepts were difficult, but because answering "is this good or bad?" required first answering deeper questions about SWMS purpose, stakeholder needs, and what "quality" actually means.
This difficulty is itself a finding: The industry is early in its AI literacy journey. We need time and space to develop informed perspectives.
Opportunities, Risks, and Uncertainties
Through station rotations, participants identified potential benefits and concerns for each AI application. What emerged was less about specific features and more about fundamental tensions.
Key Tensions Identified
Speed vs. Site-Specificity
AI can generate documents quickly, but speed incentivises generic outputs that miss site-specific hazards.
Efficiency vs. Consultation
Automated generation might bypass the consultation process that builds worker understanding and buy-in.
Standardisation vs. Judgment
Consistent outputs are valuable, but safety requires human judgment about unique circumstances.
Accessibility vs. Deskilling
Tools that make SWMS easier might reduce development of hazard identification expertise.
Opportunities Participants Saw
- Improved accessibility for small operators without safety resources
- Consistency in quality baseline across organisations
- Better compliance with regulatory requirements
- Reduction in administrative time for experienced practitioners
- Potential for learning support and hazard education
- Translation and multilingual support for diverse workforces
Risks and Concerns Raised
- Loss of site-specific knowledge and context
- False sense of security from "AI-approved" documents
- Erosion of consultation processes
- Accountability gaps when AI contributes to inadequate SWMS
- Competitive pressure to adopt tools before understanding them
- Digital divide—those without access to AI tools disadvantaged
- Over-reliance leading to deskilling of workforce
Questions We Couldn't Answer
The most valuable output might be what we don't yet know:
- How do we measure SWMS quality beyond regulatory compliance?
- What's the right balance of AI support vs. human expertise?
- Who's accountable when AI-generated SWMS miss hazards?
- How do we preserve worker consultation in AI-augmented processes?
- What training do people need to use these tools effectively?
- Do these tools improve actual safety outcomes or just documentation?
- How do we ensure equity of access to AI capabilities?
Connecting to AI Governance Frameworks
The safety profession doesn't need to invent AI governance from scratch. Established frameworks exist—we need to translate them to our context and develop practical guidance.
Relevant Frameworks for Safety Professionals
NIST AI Risk Management Framework: Provides principles for trustworthy AI including validity, safety, security, accountability, transparency, explainability, privacy, and fairness.
Australia's AI Ethics Principles: Core principles including human-centered values, fairness, privacy protection, reliability and safety, transparency, contestability, and accountability.
ISO/IEC 42001:2023 & ISO 23894: International standards for AI management systems and risk management throughout the AI lifecycle.
The Translation Challenge
These frameworks are valuable but insufficient for practitioners. Safety professionals need answers to questions like:
- What does "explainable AI" mean when reviewing a SWMS?
- How do we verify an AI tool is "reliable" for safety applications?
- Who bears accountability when AI misses a critical hazard?
- What "transparency" is required from commercial AI vendors?
Toward Simple Guardrails
Industry needs practical, actionable guidance—not just high-level principles. Workshop discussions pointed toward a "simple guardrails" approach:
Before Deployment - Ask:
- Does this serve the genuine intent of SWMS (not just compliance)?
- Have workers been consulted on the tool design and implementation?
- Can we explain how the AI reaches its outputs?
- Is accountability clearly assigned when AI-assisted SWMS are inadequate?
- Does this preserve or replace site-specific knowledge capture?
- Which stakeholder groups benefit? Which are disadvantaged?
- Can this tool fail safely?
During Use - Monitor:
- Are SWMS still being read and understood by workers?
- Is worker consultation still happening meaningfully?
- Are site-specific hazards being identified?
- Is document quality improving or just speed?
- Who's using the tool effectively vs. struggling?
Ongoing - Review:
- Are safety outcomes measurably improving?
- Where is AI helping? Where is it causing problems?
- What capabilities are being gained vs. lost?
- How is this changing safety practice and culture?
Stakeholder Roles in Responsible AI Adoption
Assuming AI becomes standard practice in safety documentation, what's the role of different stakeholders in ensuring responsible use?
The Regulator
- Clarify expectations for AI-generated safety documentation
- Define accountability standards when AI is involved
- Provide guidance on acceptable AI use cases
- Consider how enforcement approaches need to adapt
Professional Bodies (AIHS, SWA etc.)
- Develop professional guidance for AI-augmented practice
- Establish competency standards for using AI tools
- Convene ongoing industry dialogue
- Support research into AI impacts on safety outcomes
Education Providers (Universities, RTOs)
- Integrate AI literacy into safety education
- Teach critical evaluation of AI-generated outputs
- Conduct research on effectiveness and risks
- Develop continuing professional development programs
Safety Professionals
- Develop AI literacy and critical evaluation skills
- Advocate for responsible implementation
- Monitor impact on safety culture and outcomes
- Maintain core competencies in hazard identification
- Push back on inappropriate AI applications
Leadership (C-Suite)
- Make informed investment decisions about AI tools
- Ensure adequate resources for responsible implementation
- Maintain accountability for safety outcomes
- Resist competitive pressure to adopt prematurely
Technology Vendors
- Provide transparency about AI capabilities and limitations
- Engage safety practitioners in product design
- Conduct rigorous validation in safety contexts
- Support responsible implementation by clients
The Path Forward: Building Community Consensus
"We have LOADS to discuss about the potential, the opportunities and the risks that AI, especially agentic AI, will offer us. We need to work together as a community to develop responsible ways forward. That's our greatest opportunity."
This workshop revealed we're only at the beginning of understanding AI's role in safety practice.
What This Workshop Achieved
- Brought diverse stakeholders into the same room to discuss AI
- Surfaced fundamental questions about safety processes (not just technology)
- Identified tensions and trade-offs requiring ongoing dialogue
- Created shared language for discussing responsible innovation
- Documented where industry is today—early and exploratory
Immediate Next Steps
1. Continue the Conversation
- Share this report
- Invite feedback and additional perspectives
- Document evolving industry experiences with AI tools
2. Expand the Exploration
- Apply same methodology to other safety processes
- Engage different industry sectors (mining, manufacturing, etc.)
- Include workers' voices more directly
- Partner with researchers studying AI safety impacts
3. Develop Practical Guidance
- Create evaluation frameworks for AI tools
- Build case studies of responsible implementation
- Develop training resources for practitioners
- Share lessons learned—successes and failures
Beyond SWMS
This conversation extends far beyond Safe Work Method Statements. Every safety process—investigations, audits, inspections, JSAs, permit systems, training—faces similar questions about AI integration.
The methodology we used here—returning to intent, understanding stakeholder experiences, identifying tensions, connecting to governance frameworks—provides a template for exploring AI's role in any safety activity.
The Invitation
This report is an invitation to join an ongoing conversation. If you're a safety professional, researcher, regulator, technology developer, or worker interested in shaping how AI integrates with safety practice, we want to hear from you!
We're exploring questions like:
- How do we measure "quality" in AI-augmented safety work?
- What competencies do safety professionals need in an AI-enabled world?
- How do we preserve critical thinking and judgment while using AI tools?
- What does "worker consultation" mean when AI generates documentation?
- How do we ensure equity of access to AI capabilities?
- What accountability models make sense when humans and AI collaborate?
This is a call to action to build a safer, healthier and sustainable future of work.
