AI is transforming workplace safety – at least, that’s what the brochures say. The technology has the capability to predict equipment failures and surface risks that would have once gone unnoticed until someone got hurt. That potential is genuine, but so is the gap between what vendors are promising and what actually happens when these tools are deployed. Whether AI makes workers safer depends not so much on the technology than on the decisions made before it’s ever switched on.
Even so, it would be wrong to frame AI as a risk in itself. Used well, it can be a powerful ally – one that expands human capability rather than replaces it.
Twenty-five years of building technology for frontline teams teaches you that safety progress is rarely down to a single solution. It accumulates through culture, consistency and through the slow work of earning credibility with the people whose lives depend on getting it right. AI can accelerate that, but only where the foundations already exist. When it is used as a shortcut, it can put people and organisations at even greater risk.
Recent EcoOnline workplace safety research highlights this tension: over seven in ten (72%) employees say they would feel safer if their organisation used more digital health and safety tools – up from previous years – yet this confidence still hinges heavily on how those tools are implemented. At the same time, while nearly half (47%) of workers believe AI has the potential to improve workplace safety, sentiment remains cautious, with many emphasising that technology must be paired with human expertise.
This gap reveals that while appetite for digital innovation is growing, trust is still conditional and far from fully established.
That scepticism should not be dismissed, as it is born in experience of being told what technology to use rather than a shared technology journey.
Where AI genuinely strengthens safety
By nurturing a shared technology journey, in the right conditions, AI can meaningfully improve how organisations identify and respond to risk. It can analyse large datasets in seconds, helping teams spot patterns that would otherwise take weeks to uncover. It can streamline reporting, generate first drafts of safety checklists, and support faster root-cause analysis – reducing administrative burden so safety professionals can focus on higher-value work.
Some of the most effective use cases are also the simplest. During inspections or walkthroughs, AI can highlight hazards that may be overlooked in the moment, while experienced workers identify risks the system misses based on context and behaviour.
The outcome isn’t blind trust in the technology, but better discussion and better decisions. That is where AI in health and safety delivers real value – supporting human judgment, not replacing it.
There is also a broader opportunity emerging – one that echoes James Surowiecki’s idea of the “wisdom of crowds:” that better judgements can come from aggregating diverse, independent perspectives, rather than relying on a single view. While organisations can’t have multiple experts reviewing every decision, AI can act as a form of aggregated insight, drawing on patterns across teams, sites, and historical data. In doing so, it helps extend knowledge beyond the safety function and into the frontline, positioning AI as a partner to share and improve collective wisdom, rather than a replacement. When combined with true RAG (Retrieval-Augmented Generation) capabilities it takes this wisdom of crowds to another level.
The danger of “thin AI”
Nevertheless, much of what is being deployed now doesn’t live up to its billing. This is what we call ‘thin AI’: systems running on limited, inconsistent, or poorly understood data, launched without sufficient training or feedback, and sold on a demo rather than evidence of real-world performance. These tools can recognise patterns at speed, but pattern recognition isn’t the same as understanding. They can’t tell you why workers behave the way they do, how incentives shape decisions under pressure, or why trust on a factory floor takes years to build and seconds to lose.
The consequences are predictable and we see them regularly. A logistics operator introduces an AI-powered risk monitoring tool without first establishing consistent incident reporting. The model, without reliable data, flags the wrong things and misses others entirely. Workers, who were never consulted on the rollout, learn quickly that the system can’t be trusted and naturally stop engaging with it. Within months, near-miss reporting drops, not because the workplace got safer, but because people stopped feeling responsible for the outcome.
In safety-critical environments, this kind of failure is dangerous. When people defer to algorithmic outputs, human judgment atrophies. Irregularities get waved through because the system didn’t flag them, and in industries like construction, which still accounts for more workplace fatalities than any other sector, the margin for that kind of error doesn’t exist.
Frontline trust is the difference between catching something early and someone not going home.
Culture before code
Another reason why ‘thin AI’ fails is that it skips steps.
Too many organisations move directly from paper-based processes to AI-driven insights, bypassing the foundational work like building safety programmes or getting existing data into a fit state.
If frontline teams find EHS software difficult to use, they will not use them consistently. Without consistent reporting, the data feeding AI models will be incomplete or biased. Poor data leads to poor predictions, which erodes trust further. The result is an illusion of progress with dashboards filled with insights that do not translate into safer outcomes.
Our research shows that employees want better digital health and safety software, including more computer and mobile-based reporting. They want easier ways to report hazards, access guidance, and close the loop on actions taken. While mobile readiness remains a significant gap for many organisations, it remains a prerequisite for meaningful AI. Organisations must build participation first and data discipline second. Only then does AI begin to deliver real value.
Augmenting, not outsourcing, judgment
The safety leaders who get this right tend to think about AI as an additional set of eyes that can process data at a scale no human team can match, but that still needs a human judgement call.
Crucially, employees must be involved in how AI is deployed. They need to understand what the system can and cannot do, and how their feedback improves it over time. Clear feedback loops protect not only physical safety, but psychological safety. People need to feel that technology supports them, rather than replaces them.
This is also where the limits of AI need to be stated plainly. These systems do not understand fear, fatigue, or the ethical trade-offs that come with decisions under pressure.
Building resilience, not hype
AI absolutely has a role to play in the future of workplace safety.
When implemented thoughtfully, it can help organisations move from reactive compliance to proactive risk prediction, extend expertise across teams, and prevent unnecessary human harm.
But resilience is built through depth, not speed.
The question safety leaders should ask is not “How quickly can we deploy AI?” but “What must be true before AI makes us safer?”
The answer to that question does not start with technology. AI can only inherit the systems, culture, data, and trust already in place. In industries where the cost of getting it wrong is measured in lives, those foundations will always matter more than the tools themselves.
That is also where software can make its greatest impact: not as a shortcut, but as a force multiplier for the people doing the hard work of protecting others. Used responsibly, technology can help organisations make safer, faster, better-informed decisions, and play a critical role in protecting people and the planet for future generations.
