What Does Safety Train Us to Become?

Daniel and Adrian have been neighbours for almost fifteen years. Same apartment building, same generation, children who grew up around the same time. On Friday evenings they sometimes meet for a beer at the small place around the corner, where they talk about the usual things: family, football, bills, holidays, work.

Daniel works in a furniture factory. Adrian works in automotive. Both started young, both became experienced operators, both know what a serious injury can mean, and both have worked for years without a major accident. If you looked at their training records, risk assessments and occupational histories, you might describe them in almost identical language: experienced, competent, trained, reliable, safety-conscious.

But they have spent fifteen years inside two very different systems. And I have started wondering whether those systems have been doing something else during all that time. Not only protecting them. Teaching them.

Two factories

Daniel's factory has procedures. Machines are guarded, lockout matters, PPE is required, and certain rules are non-negotiable. But when something unusual happens, Daniel's supervisor has a habit of asking questions. What do you see? What has changed? What worries you? What do you think we should do?

Sometimes Daniel suggests something and the answer is simple: no, too dangerous. Sometimes the answer is show me. And occasionally: you're right, let's change the instruction. He has taken part in risk assessments. After near misses, the people who actually perform the work are brought into the discussion. When reality and procedure do not quite match, the first assumption is not automatically that the worker is wrong.

Over the years, something happens. Daniel learns the rules — but he also learns to observe, to question, to explain, to stop, to defend an opinion, and sometimes to change it.

Adrian's automotive plant takes safety just as seriously, perhaps even more visibly. There are procedures for almost everything: standard work, checklists, visual standards, authorisations, audits, sign-offs, deviation forms. When Adrian asks why something must be done in a particular way, the answer is often that it is the procedure. When something unexpected happens, the instruction is to stop and call the supervisor.

There is nothing wrong with either answer. In an environment with high-consequence hazards, tight cycle times and highly standardised work, they may be exactly the right answers — and a plant that allowed every operator to reinterpret every rule in real time would be a worse plant, not a better one. Adrian follows the instructions. He has an excellent record. He is dependable, experienced and remarkably consistent.

Fifteen years pass.

One Friday evening

The two neighbours meet for a beer, and Adrian mentions something that happened that afternoon. The machine gave an error he had never seen before.

Daniel asks what was happening.

Adrian says he doesn't know. He called the supervisor.

That was the correct response. It is what his procedure requires, and in his plant it is almost certainly the safest thing he could have done. But Daniel asks the question anyway: sure — but what did you think was happening?

Adrian pauses.

Not because he is less intelligent. Not because he is careless. Not because Daniel is a better worker. Perhaps only because he works in an environment where, for fifteen years, the system has answered a great many uncertainties before he was ever required to answer them himself.

And that leaves me with a question. After fifteen years, did these two safety systems merely produce different behaviours? Or did they help produce different professionals?

The old problem of rules and reality

This is not a new debate in safety, and that matters.

Andrew Hale and David Borys reviewed decades of research on workplace safety rules and described two contrasting paradigms. In the first, rules are largely top-down: formal, static boundaries defining what people should and should not do, with deviation treated as something to be suppressed. In the second, rules are understood as dynamic and situated; the people doing the work hold real expertise, and competence includes the ability to handle the diversity of actual situations.

Neither view can simply replace the other. Rules matter. Without them, every worker would have to rediscover safe work from scratch. Procedures preserve knowledge, standardise critical controls, reduce ambiguity and allow lessons learned by one generation to protect the next. The work of Bieder and Bourrier on proceduralisation makes this point well: routines and rules reduce uncertainty, prevent repeated mistakes and allow good practice to spread.

But they also raised an uncomfortable question. How far can proceduralisation go before its benefits begin to change character?

Research on what Rae, Provan, Weber and Dekker called safety clutter adds a second warning. Procedures, documents and activities created in the name of safety do not automatically improve operational safety. When they accumulate without a clear contribution, they have been associated with cynicism and surface compliance.

So the problem is not rules or no rules. That is far too simple. The harder question is what rules should do — and what humans should still be required to do.

Reality refuses to read our procedures

A 2026 study by Ashraf, Peres and Sasangohar examined procedural work at a petrochemical facility, using cameras mounted on workers' hard hats to compare written procedures with what operators actually did during ordinary, successful operations. Across 1,422 procedural steps, 32.9% showed some adaptation — a step taken out of sequence, performed differently, skipped, or added. Every observed task was completed successfully.

What the researchers did next is the interesting part. Rather than counting deviations, they asked experienced operators and supervisors to explain them, and then asked the workers themselves. Three functional patterns emerged, and both groups described them in almost the same words: adaptations of routine ("that's how we do it here"), adaptations of efficiency ("work smarter not harder"), and adaptations of safety ("trust but verify") — the last being workers adding verification steps that no procedure required of them.

It is a single-company study, and the authors are careful to present their framework as a hypothesis in need of testing elsewhere. But the direction is hard to ignore. Variation was not simply evidence of unreliable humans failing to follow reliable procedures. Much of it was functional adaptation to the reality of work — including adaptation that made the work safer than the document asked for.

This matters because procedures are written before reality happens. The worker meets reality when it happens. The two will never overlap perfectly. Safety therefore needs something between the procedure and the situation, and for a long time we have called part of that something human judgment.

Judgment is already part of the conversation

Martijn Flinterman recently asked a question I keep returning to: are we enabling judgment, or just enforcing execution?

Drawing on the sociologist Hartmut Rosa, he describes a wider movement from acting toward executing — from responding to situations through experience and responsibility, toward following increasingly sophisticated protocols, systems, checklists and algorithms.

The paradox is obvious. We built those systems for good reasons: predictability, consistency, accountability, fairness, risk reduction. But the question deserves to be taken seriously. If we remove too much room for judgment, do we eventually weaken adaptability, responsibility and ownership?

Safety research gives us reasons not to dismiss it. Studies in environments as tightly regulated as nuclear power have shown that empowering leadership and strong procedural compliance are not opposites. Martínez-Córcoles and colleagues found that formal procedures combined with empowering leadership clarify people's safety roles, while dialogue, participation and collaborative learning help organisations surface problems the procedures themselves never anticipated.

This breaks a false choice. We do not have to choose between compliance and thinking. A mature safety system may require both.

Sometimes safety must simply say stop. No discussion. Do not bypass the guard. Do not enter the confined space. Do not work on live equipment. Do not improvise around stored energy. There are boundaries where human discretion should become smaller, not larger — and a culture where every rule is open to negotiation eventually erodes the few that must never be negotiated.

But there are other moments where the procedure reaches the edge of what its author could anticipate, and someone still has to see, interpret, question, decide and speak.

Perhaps the question behind the question

Most of the existing discussion asks how much room for judgment a system should allow. It is an important question. But I wonder whether there is another one behind it.

Judgment may not simply be something a worker either possesses or does not. It may also be something we exercise. Something we develop. And possibly something we can gradually stop exercising.

Which takes us back to Daniel and Adrian. The question may not only be whether Adrian's system allows him to exercise judgment today. It may be what happens to a person's capacity for judgment after ten or twenty years inside a system that rarely asks them to use it.

That is a different problem. It moves us from human behaviour toward something I think safety should examine much more closely: human formation.

The system is always teaching

Think about fifteen years of work not as years, but as interactions. Thousands of them.

Follow this. Why? Sign here. What do you see? Don't question it. Something has changed — call me. What do you think? That's not your responsibility. Good catch. Just complete the form. Show me. Stop. Speak up. Wait. Think.

Each one seems almost insignificant. Repeated thousands of times, they may become something else. Because a safety system teaches even when no training is taking place. The procedure teaches. The supervisor teaches. The audit teaches. The investigation teaches. The reaction to a mistake teaches, and so does the reaction to a question. What happens after someone says I don't think this is safe teaches. And what happens when someone challenges the way things have always been done may teach more about safety than the annual training ever will.

Every safety system has a curriculum, even when nobody has written it.

What is that curriculum teaching?

Imagine Daniel and Adrian after another ten years. Both still have zero lost-time accidents, both passed every required training, both are fully competent according to the matrix.

But suppose one has gradually learned to notice weak signals, ask better questions, challenge respectfully, recognise uncertainty, help colleagues, stop when something feels wrong, explain why a rule no longer fits reality — and know when a rule must never be negotiated. And suppose the other has learned to wait for instructions, find the correct document, stay within a defined task, complete what is measured, and escalate every ambiguity upward.

Which one has the better safety record? We cannot know from this story. Possibly both.

That may be precisely the problem. Our conventional indicators tell us a great deal about what happened to them. They tell us much less about what happened inside them.

From safety performance to safety formation

I want to be careful here. This is a question worth researching, not a finding.

Safety science already studies performance, competence, behaviour, decision-making, adaptation, participation and resilience. A 2026 study by Liu, Ye, Xiang and Wang examined the match between a worker's safety decision-making capability and the demands of the task, and found something I did not expect: the relationship with decision failure is U-shaped. Failure becomes more likely when demands exceed capability — which is intuitive. But it also becomes more likely when capability substantially exceeds what the task asks of the person.

Read that second half again. It suggests that a system which consistently asks very little of a capable person is not a neutral condition. It may itself be part of the risk.

Where does decision-making capability come from in the first place? From training, certainly. From experience, individual characteristics, leadership, team learning, organisational culture. But perhaps also from thousands of ordinary encounters with the safety system itself.

Which leads me to a proposition I believe deserves investigation: human capacity is not only variable — it may also be formable.

If that is true, safety systems are not neutral containers within which people behave. Through repetition, participation, constraint, reward and response, they may gradually strengthen or weaken the very human capacities on which future safety depends: judgment, curiosity, agency, responsibility, voice, adaptability, care.

I would call this safety formation — what a safety system leaves in the person, alongside what it prevents from happening to them. I offer the name as a hypothesis, not a conclusion. What interests me is whether it can be studied.

The beer is finished

Daniel and Adrian will probably never discuss Hale, Borys, Hollnagel, Dekker, Rosa or Flinterman over their Friday beer. They don't need to.

On Monday morning they will return to their factories. The machines will start, procedures will be opened, checklists completed, supervisors will answer questions, and workers will make thousands of small decisions. And both factories will continue teaching — not through classrooms, but through thousands of small interactions about what to notice, when to question, when to stop, when to comply, when to speak, and who is allowed to decide.

Fifteen years inside a safety system may do more than give someone fifteen years of experience. They may leave something behind: a way of seeing, a relationship with authority, a willingness to speak, a capacity to judge, a sense of responsibility for the person standing beside you.

If so, we may eventually need to judge the quality of a safety system by more than the incidents it prevents or the behaviours it controls. We may also need to ask what human capacities it leaves behind.

Because a safety system does not only tell people how to work. Over time, it may teach them how to think.

And that leaves one question I cannot yet answer. If our safety systems are shaping people every day, what kind of human being are we designing?

Perhaps that is where the next part of this journey begins.