The hazard that is not on the register

From measurement to legal duty

These six models did not stay in the literature. Over the past two decades they have been built into regulation, and the obligation they now carry is not advisory.

The International Organization for Standardization published ISO 45003 in 2021, the first global standard dedicated to managing psychosocial risk within an occupational health and safety management system, addressing how work is designed, organised and managed. Australian regulators have made psychosocial risk management an explicit legal duty. The EU Occupational Safety and Health Strategic Framework 2021-2027 lists psychosocial risk and burnout among its core priorities, with national implementations across France, Germany, Italy, the Netherlands and Sweden. In Great Britain, the statutory definition of health under the Health and Safety at Work etc. Act 1974 includes psychological wellbeing, and the Health and Safety Executive’s Management Standards structure employer obligations around six areas: demands, control, support, relationships, role, and change.

Demands and control are Karasek. Support and relationships are the social dimension the Whitehall interpretation points to. What the science measured for five decades is now what the law requires employers to identify.

Article 3(2) of the Dutch Working Conditions Act (Arbeidsomstandighedenwet) requires every employer to conduct a policy aimed at preventing or limiting psychosocial workload. Article 2.15 of the accompanying Working Conditions Decree says how: the relevant risks are recorded in a statutory risk inventory and evaluation (the RI&E), and the measures in a plan of action. The Act names five themes under psychosocial workload: discrimination, sexual harassment, aggression and violence, bullying, and work pressure. The Netherlands Labour Authority enforces this and names work pressure and undesirable behaviour explicitly in its 2026 programme (Nederlandse Arbeidsinspectie, 2025).

The scale is not marginal. Psychosocial workload is reported as the most common occupational illness in the country, and it accounts for more than half of what work-related sickness absence costs Dutch employers in continued wage payments: 4.9 of 8.3 billion euro in 2023. It takes that share because these absences run longest. Someone signed off with psychological complaints, overstrain or burnout is away for around 63 working days, and in three out of four of those cases the employee says the work is at least part of the reason. The share is also rising: employees reporting burnout complaints went from 13% in 2015 to 19% in 2023 (TNO, 2025).

So the position across many jurisdictions is now this. Employers have a duty to identify psychosocial hazards, assess them, apply controls, and monitor the result. The instrument is a hazard register. The logic is the one that has worked for physical safety for a century: name the hazard, and the control follows.

What the register can hold

Four of those five statutory themes are attributional in form; one is conditional.

Discrimination, sexual harassment, aggression and violence, and bullying each require an identifiable person doing something they should not. Each generates an incident: a moment, a date, an account, and someone who can be named. The register was built around exactly this, because it was built for physical safety, where hazards can be pointed at and incidents are things that happen.

Work pressure is the one conditional theme, and it is unattributed: a condition with no named source, measured by volume — too much work, too little time.

What DFS names is on neither list. A choice that binds without its reasoning being given to the people it binds is not discrimination, harassment, aggression or bullying, because it requires no hostile actor and typically has none. And it is not work pressure, because the problem is not volume. Robert’s difficulty was never that there was too much work. It was that the work had been shaped by choices he could see were wrong and had no way to question.

So the most prevalent psychosocial hazard in these organisations is invisible to the instrument that is legally required to inventory it. Not because anyone is evading the duty. Because the hazard does not have the shape the register can hold.

There is no incident to enter

The reason follows directly from the architecture, and it is the same property that makes the harm so hard to describe from inside.

Where conduct is prohibited, breaching the prohibition produces an event: something happened, on a date, that should not have. Where nothing was ever required, there is nothing to breach. No moment when the reasoning was refused, because it was never owed. No refusal to record, no meeting that went wrong, no words anyone said.

So the person harmed this way has nothing to report. Not nothing to prove: nothing to report. What accumulates over years is not a sequence of events but the absence of one: all the occasions on which perceptiveness could have been drawn out and was not.

This may well be why Robert’s own answer to “why did this happen to me” is “bad luck”. Bad luck is what is left when harm is real and there is nothing to point at. The invisibility follows from the architecture itself, and the hazard register is where it can be shown: the column that would hold this hazard is a column for incidents, and this hazard produces none.

Why the interventions that follow do not work

Where the hazard cannot be identified, the duty gets discharged with what is available, and what is available is aimed at the person rather than the condition. Resilience training. Mindfulness provision. Stress management workshops. Employee assistance programmes. Wellbeing apps.

Fleming (2024) tested this at scale, comparing participants and non-participants across 46,336 workers in 233 UK organisations. Across multiple wellbeing indicators, participants appeared no better off than those who did not take part; resilience and stress management showed some indication of harm. His conclusion is that organisations have to change the workplace and not just the worker.

DFS supplies the mechanism the finding leaves open. These interventions operate on the person’s response to the condition. The condition is untouched, and the person returns to it on Monday. That is not a shortcoming of the intervention. It is what an intervention aimed at a response does when the thing producing the response is still running.

What a control would specify

Under a duty to identify and control, an argument of this kind is only useful if it produces something a register can hold and an inspector can check. It does.

The hazard is identifiable in structural terms: choices that bind others, which the organisation does not require anyone to explain and answer questions about before the people they bind have to act.

The control is specifiable: choices that bind others have to be substantiated before they bind. The authority holder substantiates the choice. Or they declare it a decision, and have the risk it carries managed like any other. Or they draw out the collective perceptiveness of those involved, and approve what has been substantiated.

Specifying a control is not the same as putting one in place. What a specification does is make the hazard nameable and the control checkable. Under a duty of this kind, that has to come first.

And the control is verifiable, which is what matters for a register. Where the requirement is in place, choices that bind leave a trace: reasoning that was given, questions that were asked, limits that were declared before the work began. Without it, there is no trace. Whether it is there is observable without asking anyone how they feel.

Every existing instrument for psychosocial risk assessment measures perception: how burdened people report feeling. This control is checkable against what the organisation does when a choice binds. It does not depend on anyone being willing to say, in a survey, that their manager cannot answer questions.