As AI moves from experimentation to enterprise deployment, progress in workplace safety has plateaued.
According to new data from the UK Health and Safety Executive (HSE), 126 workers were killed in work-related accidents in Great Britain during 2025/26—two fewer than the previous year. While the long-term trend may be small steps in the right direction, many of the most serious risks have persisted for years.
AI has the potential to reduce serious harm figures in ways that previous efforts could not. But not in the way you expect.
Safety’s Tendency to Think About AI Too Narrowly
Most conversations about AI in environment, health and safety (EHS) today focus on how automation can reduce safety teams’ administrative load. How can AI write reports faster, summarise meetings and answer questions from the field?
These applications matter because safety teams spend significant time managing documentation, producing reports and consolidating information. According to a 2026 study conducted by the What Works Institute, most organisations are still in the early stages of exploring AI for EHS, using it for reporting, training support and incident investigations.
Automation speeds up tasks, improves accuracy and allows teams to spend more time away from their desks, gaining valuable insights while talking to frontline workers. These are important first steps.
But productivity alone won’t fundamentally change safety outcomes.
Emerging Workplace Risk
The greater opportunity for AI lies not in automating tasks but in helping humans make sense of increasingly complex operational environments and pinpointing emerging risks.
Modern organisations generate enormous volumes of safety data. Training records, risk assessments, inspections, incident reports, near misses, audits, corrective actions, contractor information and equipment maintenance logs are a few, and they sit in siloed systems.
The challenge isn’t a shortage of information. It’s interpreting it in a timely manner.
With data across multiple systems, teams and workflows, risk signals are fragmented. Viewed individually, many of these records appear insignificant. Viewed collectively, they can reveal patterns that point to emerging hazards.
Humans are exceptionally good at applying judgment and context. What we struggle to do is connect thousands of weak signals across large datasets in real time. As operations become more complex, the gap between available information and actionable understanding grows.
This is where AI begins to make a real difference.
How AI Uncovers Hidden Risk
The most valuable applications of AI in workplace safety are not necessarily predictive. They are interpretive.
AI excels at reviewing information holistically rather than sequentially. It can analyse incidents, inspections, audits, observations and corrective actions together to identify relationships and root causes that might otherwise be overlooked.
This shift matters because serious workplace incidents rarely result from a single event.
Rather than examining a single report in isolation, AI evaluates the broader context.
- What task was being performed?
- What hazards were present?
- Were controls missing or degraded?
- Have similar events occurred elsewhere in the organisation?
By synthesising fragmented information at scale, AI can surface patterns humans may overlook and highlight combinations of factors associated with elevated risk. This doesn’t replace professional judgement. It strengthens it.
Data Readiness Comes First
AI can’t identify meaningful risk from unreliable information. To fully benefit from AI, organisations need a strong data foundation.
According to McKinsey & Company, there’s a stark divide between successful AI programs and those that stall. Successful AI implementations come from organisations that spend roughly 70% of their budgets on data readiness. Less successful peers spend less than 30%.
Here are five foundational data principles to consider for a resilient, intelligent safety framework.
- Start with clear objectives
Safety goals should align with your organisation’s broader objectives and clearly outlining how data will support these goals will help you get there.
- Measure what matters
Select KPIs and metrics suited to your industry, organisation size and specific risks.
- Collect data consistently
Implement a standardised system that captures relevant safety information consistently. Tools like incident reporting software, wearable devices or mobile apps help collect data.
- Continuously improve data quality
Regularly audit your data, cross-check information and review reporting processes and address discrepancies promptly.
- Share information openly
Create channels to share safety data with stakeholders and provide accessible reports that monitor safety performance. Promote a culture of transparency by encouraging open discussions on safety issues.
Governance is How We Get There
AI can help identify patterns, synthesise information and surface weak signals earlier than humans can at scale. But humans remain responsible for making decisions, evaluating trade-offs and implementing action.
Trust, explainability and accountability remain essential.
Companies that establish clear decision rights, transparent review processes and strong oversight will gain the greatest value from AI.
The organisations that realise the greatest value from AI won’t necessarily be those that automate the most work. They’ll be the ones that combine trusted data, thoughtful governance and human expertise to recognise risk earlier.
