What is safe?

The work safety profession spends a great deal of effort defining safety and almost none interrogating the word “safe”. Yet safe is the word that does the work when it comes to work safety. It is the word a worker uses before climbing the scaffold, the word a board hears in an assurance report, and the word a regulator reaches for when something has gone wrong. Drawing on the recent empirical finding that safety professionals across North America hold no shared definition of safety (Raheemy et al. 2025), this paper turns from the noun to the adjective and argues that safe is not a description of the world but a judgement we pass on it. Read through philosophy of language, risk science, organisational sociology and the Australian statutory duty, the word reveals at least eight ways in which it is routinely misconstrued such as a property rather than a verdict, as an absolute rather than a comparative, as a presence inferred from the absence of its opposite, as a single state rather than three different temporal claims, as control rather than luck, as feeling rather than fact, as one meaning where several are in play, and as a legal standard it never was. Scaled up to the level of the state, the same features make the word a potent political instrument (as the pandemic language of keeping us safe illustrates),  to govern in safety’s name is to draw on the oldest justification for state power while placing a contestable judgement, and the trade-offs it hides, beyond easy challenge. The same leverage runs through the workplace itself, where unsafe can justify a stoppage and safe can press the work onward, a tool available to union and employer alike. The argument therefore is not that the word should be abandoned but rather it is that the honest form of the word is conditional and comparative, and that a profession willing to say safer, for whom, from what, judged acceptable by whom, and over what horizon would communicate better, decide better, and mislead itself less.

This article is a reprint of the article first published by the author September 2025.

The word we trust most and examine least

Ask a roomful of experienced work safety professionals what safety means and you will get a roomful of confident, incompatible answers. That is simply not speculation, it’s the authors lived experience. Furthermore it is the central finding of a careful piece of empirical work. Raheemy, Sherratt and Hallowell (2025) put a single question, ‘what is safety?’, to 518 professionals across commercial construction, utilities, oil and gas, infrastructure and related sectors in North America, and analysed the responses thematically. No dominant definition emerged. People were passionate about their own understanding, but not one understanding was widely shared (Raheemy et al. 2025). The authors organised the divergence along a temporal axis: some defined safety as something about the future, some about the present, and some about the past.

This paper takes that finding as a starting point and changes one word in the question. Not ‘what is safety?’ but ‘what is safe?’ The shift looks small but comparatively is not. Safety is a noun, and treating it as a noun invites us to look for a thing, a state, a discipline, a quantity that might one day be measured. Safe is an adjective, it’s the the thing we actually assert of a person, a place, a tool, an activity or a system when we make a decision about it. The operative sentence in most workplaces is never ‘discuss the ontology of safety’. It is ‘is this safe?’ or, more honestly, ‘is it safe enough to start the job?’

We all use the word so often, and apparently so successfully, that we assume it means the same thing to everyone who hears it. A site manager declares an area safe. A worker reports feeling safe. A designer certifies a guard as safe. An executive tells a board the operation is safe. Each of them uses the same word, and each assumes the others understand what they mean by it. This assumption is seldom tested, because the word seems too plain and simple to need testing (Ball and Frerk 2015; Hollnagel 2014b). The author’s argument is that this plainness of the word safe is an illusion, and a sometimes expensive one. Safe is among the most loaded words in the working vocabulary precisely because it looks like one of the least.

What follows in this article is a philosophical examination of the use of the word rather than a literature review, though it leans on a wide literature to make its case. The author wants to show that safe is not something we discover about the world but something we decide about it, that decision sometimes hides the values we seldom name, and that much of the confusion, miscommunication and self-deception in workplace safety can be traced to forgetting this single fact.

The grammar and the history of the word “safe”

The history of the word is worth a small moment, because it records a migration of meaning that still shapes how we argue it today. The lexicographer Douglas Harper (2001) traces safe into English around 1300, from the Old French sauf, itself from the Latin salvus, meaning uninjured or healthy. The Latin root reportedly sits in a family that includes salus, good health, and solidus, solid, all reaching back to a Proto-Indo-European sense of wholeness. The earliest idea of being safe was therefore inwardly. It described the condition of a person’s body, their soundness, their not-being-harmed. Only later, in the 1580s, did safe take on its adjectival use for conditions and situations described as being without risk (Harper 2001). With that move the word turned outward. The word stopped being only about the state of a person and started being about the state of their surroundings.

That migration of the word safe, from a property of a person to a property of a situation, matters because it set up the modern habit of speaking as though safe were a feature of the world that we inspect, the way we might inspect a wall for cracks. Raheemy et al. (2025) note the same trajectory, observing that contemporary discussion has shifted again, this time toward the system rather than the individual as the place where safety is said to happen. There are therefore three locations for safety, then, layered across the centuries. Specifically, the body, the environment, the system.

The grammar of the word appears more revealing than its history. The philosopher J.L. Austin, in Sense and Sensibilia (1962), drew attention to a class of words he called “trouser-words.” A trouser-word is one word whose meaning is carried by its negative. We do not learn what real means by grasping some positive essence of reality in that we learn it by contrast with the fake, the decoy, the imitation. As Austin put it, with real it is the negative use that wears the trousers (Austin 1962b). The positive term is a placeholder that likely means little more than ‘not subject to any of the ways of being unreal’.

Safe appears to behave the same way. We do not, typically in practice, possess a positive specification of what makes a thing safe. What we possess is a long and detailed vocabulary of  unsafe such as the hazard, the defective equipment, the near miss, the breach, the non-conformance, the incident. We often recognise danger far more readily than we recognise its absence. A work safety inspection is in truth an “unsafety inspection”, a search for the ways a situation falls short. When the search finds nothing, we call the situation safe, which is to say we have failed to find it unsafe. The word is doing the work of a double negative dressed up as a positive claim if you can believe it. This is not a quibble about language, rather it explains why so much of the discipline is organised around finding what is wrong, and why declaring something safe so often means no more than ‘we looked and did not see a problem today’.

Safe is a judgement, not a property

If the grammar tells us that safe lives off its opposite, the philosophy of risk tells us something more unsettling in that even where we can specify it, safe is not a fact about the world at all. It is ultimately a value judgement we make about the condition of the world.

From the author’s research, the clearest statement of this remains one of the oldest. William Lowrance, writing in 1976, proposed that a thing is safe if its risks are judged to be acceptable (Lowrance 1976, p. 8).  Judged tells us that safety is the output of an act of evaluation by some person or body, not a measurement returned by an instrument. Acceptable tells us that the evaluation rests on a standard of acceptability that is itself a choice, and a choice that different people, with different stakes, will make differently. Lowrance’s formulation does not describe a property of an activity, it ultimately describes a verdict passed upon it.

The major reference points in risk science have likely converged on this view, even as they argue tirelessly about the minor details. The Society for Risk Analysis glossary, assembled by a broad community of experts, defines safe as being without unacceptable risk and treats safety as the antonym of risk, with high safety corresponding to low risk (SRA 2018). The international standards community takes the same line that the ISO/IEC framing presents safety as freedom from risk that is not tolerable (ISO/IEC 2014). In each case the apparently objective word safe turns out to depend on a prior judgement about which risks a society, an organisation or an individual is prepared to live with. Möller, Hansson and Peterson (2006) press the point philosophically in a paper whose title states it plainly “safety is more than the antonym of risk, and cannot be reduced to a low probability of harm alone.” Aven (2014) makes a parallel case from risk science, warning that treating safety as the bare opposite of risk is too thin, because it leans on probability alone and ignores the deep uncertainties of real settings.

Once we accept that safe encodes a judgement of acceptability, an awkward question must ultimatly follow immediately. Acceptable to whom you may ask? The answer is rarely the same for everyone in the room as the author previously eluded to. The risk that an experienced rigger accepts without a second thought may be intolerable to the apprentice working beside him, to his family waiting at home, to the workplace insurer, or to the work safety regulator reading the file after the fact of an incident. Unfortunately, there is no neutral vantage point from which acceptability can be deduced from. This is why the principle of reducing risk to a level that is as low as reasonably practicable, and its variants, had to be invented at all. They exist to give the notion of acceptability a defensible procedure, a way of showing the working behind the verdict (Jones-Lee and Aven 2011). The procedure is highly valuable but what it cannot do is convert a judgement into a fact. The threshold of acceptable risk remains, as Raheemy et al. (2025) put it in reviewing the same literature, somewhat elusive.

When a manager at a work site says an activity is safe, the honest translation is ‘I judge the residual risks of this activity to be acceptable, by a standard I am applying and you may not truly share.’ Stated that way, the claim invites the right next questions,  acceptable by what and who’s standard? Acceptable to which of the people exposed to the hazard? Stated as a  ‘it’s safe’, the claim shuts those questions down and disguises a value choice as a finding of fact, which is fundamentally the problem at hand.

Safe is likely graded but never absolute

There is a second, related confusion, and it concerns whether safe names a state you can fully reach or a direction you can only move along.

The everyday use of the word treats this as binary that a thing is safe or it is not, the way a switch is on or off. The literature is close to unanimous that this binary is likely a false assumption. Residual risk is a feature of every activity that has not ceased entirely; reduce a risk as far as you like and something always remains short of stopping the work altogether (Manuele 2013). Manuele argues bluntly that describing safety as the reduction of risk to zero betrays a misunderstanding of the relationship between risk and safety, because the zero state is unattainable while any activity continues. The implication is that ‘completely safe’ is not a demanding standard, it is likely an incoherent one. Unfortunately, that has not stopped ‘zero harm’ and Zero Accident Vision from becoming influential commitments across many industries, where zero functions as an aspiration and a statement of values rather than a literal claim about any present state, a stance its proponents defend and its critics question (Zwetsloot et al. 2017). That’s a conversation for another day.

Möller, Hansson and Peterson (2006) capture the structure of the confusion by distinguishing two concepts that the single word safe is made to carry. There is an absolute sense, in which safe means the complete absence of risk, and a relative sense, in which safe means risk reduced to some acceptable level. The two are not refinements of one idea, they are different ideas wearing the same word. Trouble arises whenever a conversation slides between them without anyone noticing. For example when a worker who hears ‘this is safe’ in the absolute sense, and is then injured by a residual risk that management  or someone else in the chain had judged acceptable in the relative sense, has not been lied to exactly. They have likely been failed by a word that let two meanings pass for one.

The honest form of the word, in almost every real case, is comparative rather than absolute.  Not that a thing is safe, but that it is safer than the alternative form. We can compare two methods and say with confidence which carries the lesser risk. We can compare a worksite this month against the same site last month and say whether the controls have improved. What we cannot do, without tacitly importing a contestable threshold, is plant a flag at some point on the scale and declare everything beyond that flag is safe. The comparative claim is modest and can be defended and the absolute one promises more than any continuing activity can deliver, and much of the rhetoric of workplace safety prefers it anyway, with consequences the next section takes up.

The confusion was, is, and will be

The most useful structural insight in Raheemy et al. (2025) is that the divergent definitions of safety sort themselves along the essence of time itself. Some respondents located safety in the future, defining it as the actions taken so that work will be safe preventing incidents, managing risk, managing hazards, improving continuously. Others located it in the present, as the condition of being protected and the presence of controls right now. A small minority located it in the past, as the absence of harm, an outcome you can only read once the work is done. In the data analysed by the authors, the future-facing themes were the largest, with prevention of incidents the single most common at 24 per cent and hazard management close behind at 19 per cent; the present-facing condition of being protected drew 21 per cent; and the past-facing absence of harm drew only 2 per cent (Raheemy et al. 2025).

Turn that finding on the adjective and a specific failure of communication comes into focus.

When three people say a job is ‘safe’, one may mean it will be safe because the controls are planned, one may mean it is safe because the controls are in place, and one may mean it wassafe because no one got hurt. These are not three shades of one claim they are three different claims about three different moments, fundamentally joined only by a shared word. A supervisor reporting present control and an executive recording a past outcome can sit in the same meeting, use the same sentence, and mean things that do not link to one another at all.

The past-tense reading is the most dangerous of the three, and it is worth focussing on, because it is where the harm is hardest to see. Defining safe as the absence of harm makes safety an outcome, and an outcome you can only observe after the fact. Hollnagel (2014a) describes this as treating safety as an epiphenomenon, a by-product visible only through the non-occurrence of accidents. The trouble is that you cannot study, manage or measure a non-event directly in that there is nothing there to observe. Reason (2000) made the same point years earlier, noting that counting the instances of harm that did not happen is not a coherent measurement task. What we end up studying, when we define safe as the absence of harm, is the absence of safety rather than safety itself (Vandeskog 2024).

This is not an abstract, it produces a specific and well-documented error that inferring that work was safe from the fact that no one was hurt. The inference is likely invalid because it confuses a good outcome with a sound process. Work can be performed in a thoroughly unsafe manner and produce no injury, in which case the absence of harm is owed to luck, not to control (Raheemy et al. 2025). Hallowell and colleagues (2021) demonstrated the statistical version of this failure, showing that injury-rate metrics such as the total recordable incident rate are so unstable that they are effectively meaningless for most business decisions or research. A period without recordable injuries is often read as a period of safety but the reading likely has no statistical reliability for the future.

Organisational sociology supplies the deeper mechanism, and it has a specific name. Vaughan (1996), in her study of the Challenger launch decision, called it the normalisation of deviance “the gradual process by which a practice that was once recognised as a departure from the standard becomes, through repetition without catastrophe, accepted as normal”. Each flight that returned with eroded O-rings made the next launch in cold conditions feel a little more acceptable, until a known danger had been quietly reclassified as routine. Rasmussen (1997) modelled the same dynamic at the level of the whole socio-technical system, describing how organisations under pressure to be efficient and to reduce effort migrate steadily toward the boundary of safe operation, with each local decision reasonable on its own terms and the drift invisible from inside. Snook (2000), examining the accidental shooting down of two US Army helicopters by friendly aircraft, named the version that lives in the gap between rules and work ie the practical drift, the slow decoupling of the procedures on paper from the practices on the ground. Dekker (2011) gathered these threads into the broader idea of drift into failure, the proposition that disasters in complex systems often emerge not from broken parts but from normal people doing normal work as the system slides, unremarked, toward the edge.

The common thread is that ‘it was safe, because nothing went wrong’ is precisely the sentence that licenses the drift. Every uneventful repetition of a degraded practice is taken as confirmation that the practice is safe, and the confirmation pushes the boundary a little further. The word safe, used as a verdict on the past, becomes the instrument by which an organisation talks itself into riskiness and danger.

Philosophy potentially also offers a clarifying mirror here, from an unexpected quarter. In epistemology, the word safe already has a precise technical meaning, and it is not the everyday one. Following Sosa (1999), and developed by Williamson (2000) and Pritchard (2005), a belief is said to be safe not when it happens to be true but when it could not easily have been false, when it would remain true across the nearby ways things might have gone. Safety, in this account, is a modal property such as robustness across close possibilities, not mere success in the actual case. Carry that intuition back to the workplace and the outcome-substitution error stands exposed. An activity that produced no harm once is not thereby safe, in the demanding sense, if harm would easily have occurred under a slightly different operator, a little more fatigue, one missed step or a colder morning. Real safety is likely robustness across those nearby variations. A single good outcome tells you almost nothing about it and if it can be repeated. The analogy is not an argument by itself, but it sharpens the point that safe worth the name describes how the work holds up across the conditions it might have met, not the single condition it happened to meet.

Felt safe and being safe

So far the author’s argument has treated safe as an objective judgement, however contestable its standard,  but the word likely has a thoroughly subjective life as well, and the two lives are constantly mistaken for each other.

People do not only assess safety, they typically feel it. The sense of being safe is a psychological state, shaped by familiarity, by trust in colleagues, by the visible presence of equipment, and by a great deal that has nothing to do with the actual distribution of risk. Slovic’s long program of work on the perception of risk established that lay judgements of how dangerous something is track factors such as dread, control and familiarity far more than they track measured probabilities (Slovic 1987). The feeling of safety and the state of safety can therefore come apart in both directions. A worker can feel perfectly safe while standing in real danger, because the danger is familiar and unseen. A worker can feel deeply unsafe while objectively well protected, because the hazard is novel or dramatic. Each gap matters because the first kills people who never saw it coming and the second drives resistance to controls that are actually sound.

The international effort to define safety has also stumbled on exactly this split. When two World Health Organisation collaborating centres worked toward a shared definition over a two-year consensus, they deliberately built in both dimensions, framing safety as a state in which hazards are controlled to preserve health and wellbeing, and explicitly including the subjective perception of safety alongside the objective control of hazards (Maurice et al. 2001). The definition was likely sound but its uptake was unbalanced. Injury-prevention programs and researchers gravitated to the measurable, objective side, reducing safety to falling injury rates and dropping the perceptual dimension the definition had been careful to include (Raheemy et al. 2025). The richer concept was available, the narrower one was easier to count, so the narrower one typically won the arguments.

There is a further twist that turns safe from a description into an aspect of performance. We do not only use the word to report a state, we use it to reassure ourselves. ‘You’ll be safe today – nothing to worry about here’ said to a new worker is rarely a calibrated risk statement. It is a speech act whose function is comfort the worker, the verbal equivalent of a steadying hand on the job. There is nothing wrong with reassurance in its place, but the danger is that the comforting use and the descriptive use share a word, so that a soothing ‘it’s safe’ can be heard, and later relied upon, as a considered finding. When the word is asked to console and to assess at the same time, the assessment is the part that quietly disappears.

There is also an entirely separate sense of the word, now everywhere in workplaces across Australia and the world, which describes neither physical risk nor its perception. Edmondson (1999) introduced the construct of psychological safety, defined as a shared belief that a team is safe for interpersonal risk-taking, safe to speak up, to admit error, to ask the awkward question without fear of humiliation. This is a precise and valuable idea, and it has nothing directly to do with whether the scaffold will hold and not fall. Yet it travels under the same adjective. A workplace can be physically dangerous and psychologically safe, or physically well controlled and psychologically toxic. When a survey reports that a team feels ‘safe’, the word may be pointing at candour and trust, or at the absence of physical hazard, or at a feeling about physical hazard, and the three are routinely conflated in the same report. By now the single word safe is doing at least four different jobs.  The controlled state, the feeling of that state, the reassurance about it, and the social permission to speak.

Why agreement about ‘safe’ is elusive and perhaps impossible

It is tempting to read the lack of consensus that Raheemy et al. (2025) documented as a problem to be solved, a sign that the profession simply has not tried hard enough to settle on a definition but there are two strands of philosophy suggest the difficulty runs deeper than effort.

The first is Wittgenstein’s idea of family resemblance. In the Philosophical Investigations (1953), Wittgenstein observed that many of our most-used concepts are not held together by a single feature common to all their instances. Instead they are held together by an overlapping web of similarities, as the members of a family share a look without any one feature being present in all of them. Game is his example eg there is no property shared by chess, football, solitaire and a child’s make-believe, only a network of resemblances. Safe looks very much like such a concept. The future-facing, present-facing and past-facing uses, the felt and the controlled, the physical and the psychological, share overlapping family features without a single essence running through all of them. If that is right, then the search for one precise definition that captures every legitimate use is not merely hard,  it is fundamentally looking for something that is not there.

The second is Gallie’s idea of essentially contested concepts. Gallie (1956) identified a class of concepts, among them democracy, art and social justice, whose proper use is the subject of endless and genuine dispute, not because the disputants are confused but because the concept carries an internal value component about which reasonable people will always disagree. The disputes are not resolvable by gathering more facts, because what is at stake is partly a question of what we ought to value. Safe fits the pattern closely. Because the word encodes a judgement of acceptable risk, and because acceptability is irreducibly a question of values and of who bears the cost, disagreement about what counts as safe is not a temporary state of immaturity in the field. It may be a permanent feature of the concept itself.

This is not a counsel of despair, and Raheemy et al. (2025) are likely right to insist that the divergent understandings are complementary rather than mutually exclusive. The future, present and past readings of safety describe genuinely different and connected moments such as the inputs we plan, the conditions we maintain, and the outcomes we record. A mature account holds all three together rather than choosing one. But holding them together is a very different project from collapsing them into a single definition, and it is worth being honest about which is achievable. We are unlikely ever to agree on one sentence that fixes the meaning of safe for all purposes and all parties but we can, far more usefully, agree to be explicit about which sense we are using each time we use it.

“Safe” in the mouth of the state

The reassurance use examined earlier in the article has a public form, and it is the most powerful one the word has. When an individual says “you’ll be safe”, the act is small and local. When a government says “we will keep you safe”, the same act is performed at scale of the state, and it likely draws on the oldest justification for state power there is.

That justification is an ancient story. Cicero, setting out how a republic ought to be governed, put it as a maxim: salus populi suprema lex esto, let the safety of the people be the supreme law (Cicero, De Legibus, 3.3.8). Hobbes placed the same principle at the head of his account of the sovereign’s office, making the safety of the people the very business for which the sovereign exists (Hobbes 1651, ch. 30). The idea has run through Western political thought ever since in that the first thing a state is for is to keep its people safe. To speak of keeping the people safe is therefore not an ordinary political promise but it reaches the foundation on which the legitimacy of government rests which is exactly what makes the phrase so powerful and so hard for us to argue against.

It is also, for reasons this paper has already set out, a slippery, slippery slope. Austin, whose account of the negative life of words appeared earlier, devoted a separate book to utterances that do not describe the world but act upon it (Austin 1962a). “We will keep you safe” is one of these. It does not report a state of affairs; it performs an act, casting the speaker as protector and the listener as protected, and inviting trust rather than scrutiny. The international relations scholars of the Copenhagen School built directly on Austin to describe a move they called securitisation. To name something a matter of safety or security, they argued, is itself a speech act, and once an audience accepts the framing, the issue is lifted out of normal politics into a state of exception, where measures that would otherwise be debated become urgent necessities and to question them can look like siding with the threat (Wæver 1995; Buzan, Wæver and de Wilde 1998). Those same scholars warned that this is a power to be used sparingly, precisely because securitising an issue tends to bypass the ordinary deliberation through which a society weighs costs against benefits.

Everything established earlier about the word explains why the move works. Safe is a trouser-word, so “keeping you safe” defines itself against a danger that need not be named and a standard that need not be stated, which leaves the claim almost impossible to falsify. It sounds absolute, promising a condition no continuing activity can fully reach. And it conceals the judgement at its centre: when a government declares that a measure will keep people safe, it has already decided which risks count, what level of them is acceptable, and which competing harms are to be set against them, all folded into a single word that names only one side of the ledger and never reveals that anyone is holding the scales.

The pandemic is the clearest recent illustration, and the analytical point can be made without taking any position on whether the measures were warranted or not. Governments around the world including Australian federal and state governments among them, framed their response in the language of safety, urging people to stay safe, to keep the community safe, and to adopt COVID-safe behaviours (Australian Government Department of Health and Aged Care 2022; NSW Health 2023). On one side of the ledger sat the virus and on the other sat lives and livelihoods, schooling, mental health, deferred medical care and the isolation of the old and the dying. To be for safety was to be for the measures to control the risk, and to question a measure could be heard as being against safety in itself. The bare predicate did not perform the weighing of those competing harms. It concealed that a weighing was taking place and that particular people were doing it, on the public’s behalf, against a threshold the public was rarely shown. None of this implies bad faith at all rather a government that invokes safety may be entirely sincere, and the harm it points to may be grave. The point is likely structural in that the word does this work whatever the intention behind it, because of what the word is.

The line most often quoted to resist this kind of talk has an irony in its history. “Those who would give up essential Liberty, to purchase a little temporary Safety, deserve neither Liberty nor Safety”, wrote Benjamin Franklin on behalf of the Pennsylvania Assembly in 1755 (Franklin 1755).

The sentence is now a slogan for the primacy of liberty over security, yet in its original setting it concerned a colonial tax dispute, in which the “liberty” at stake was a legislature’s power to levy taxes and the “safety” was a wealthy family’s wish to be exempted from them. Even the words we reach for to scrutinise safety are themselves routinely misremembered, which is only further evidence of how unstable this vocabulary is and how much it rewards the discipline this paper has urged.

The honest political claim takes the same form as the honest claim anywhere else in that it is comparative and conditional. Not “this will keep you safe”, but “this measure reduces this risk, at these costs to these people, and we judge that trade-off acceptable for these reasons”. Put that way, the questions the bare word suppresses, safe from what, for whom, at what price, and judged acceptable by whom, are returned to public view, which in a democracy is where they belong.

Safe and unsafe in the industrial contest

The previous section we considered how the state wields the word “safe”. But the same leverage operates inside the workplace, between those who employ labour and those who sell it, and it runs in both directions.

Begin with the lawful core, because the argument that follows is not a challenge to it. The model Work Health and Safety Act gives a worker a genuine and important right to cease, or refuse to carry out, work where the worker has a reasonable concern that doing so would expose them to a serious risk from an immediate or imminent hazard (Work Health and Safety Act 2011, s 84). It gives a trained health and safety representative the power to direct that unsafe work cease in the face of such a risk (Work Health and Safety Act 2011, s 85), and it protects workers and their representatives from discrimination for exercising these rights (Work Health and Safety Act 2011, ss 104 to 105). These protections were hard won, they prevent real harm, and nothing here is meant to weaken them. They are the legitimate use of unsafe, and they are the baseline against which any misuse has to be measured.

Misuse is possible because of the very features this paper has described. Unsafe, like safe, resists falsification in the moment, and it carries a moral authority that is hard to answer, because no one wishes to be heard arguing against safety. Those two properties make the word available as a lever. A claim that work is unsafe can be attached to a dispute whose real driver is something else, bargaining position, demarcation, or a wider industrial campaign, and once the word is in play the genuine concern and the tactical one can look alike from the outside, because the word names only the safety side and conceals the rest. In Queensland this possibility is currently before a public inquiry. The Queensland Commission of Inquiry into the CFMEU and Misconduct in the Construction Industry (2026) is examining claims of misconduct connected to the union and the wider construction industry, including a decade of industrial disputation. These are allegations the inquiry has yet to resolve, and the union rejects them, stating that it never compromises on safety and that it is being scapegoated to justify a broader industrial agenda (CFMEU Queensland and Northern Territory 2025). The merits are not for this paper to settle, what matters here is that each side casts the other as the one using safety, or the politics of safety, as cover for ends that are not about risk, which is as clear an illustration as the argument could want that the word has become a contested instrument.

The same word is turned the other way by employers, and an honest account must give that weight as well. Where a worker hesitates, an unscrupulous employer can use ‘it’s safe’ as the reassurance act examined earlier, less a finding than a manoeuvre, to override the hesitation and move the work along (of course there are severe penalties that may apply now). The right to cease unsafe work can be recast as overreaction, and the worker who exercises it painted as obstructive. A contested situation can be declared safe by unilateral assertion, which fixes the threshold of acceptable risk in the employer’s own commercial interest while hiding that any threshold has been set. Schedule and cost pressure supply a standing motive for precisely this, the same pressure behind the slow migration toward the boundary of safe operation that Rasmussen (1997) described. Here too the statute sets limits in that coercing a worker, misrepresenting a matter of health and safety, or penalising someone for raising a safety concern is itself prohibited (Work Health and Safety Act 2011, ss 104, 108, and 109). The worker pressed back onto the job by a confident ‘it’s safe’ has been failed by the same word, used in the same way, as the worker kept off it by a strategic ‘it’s unsafe’.

The symmetry is the fundamental lesson in point in that the manipulation of safe belongs to no single side of industry. It is available to whoever has an interest, because of the structure set out across this paper that a trouser-word that resists falsification, an absolute ring that promises more than any activity can deliver, a concealed judgement of acceptability, and a moral charge that makes opposition feel indefensible. The discipline that answers misuse is the same in both directions. A genuine claim survives specification, safe or unsafe from what, for whom, on what evidence, and against which standard, while a tactical one does not. Asking those questions protects the honest worker and the honest employer alike, because it is exactly what separates them from the versions that trade on the bare word. It is also close to what the law already does, as the next section shows, since the statutory duty is framed not around the word safe but around a qualified obligation to weigh the risk and to act.

What “safe” means for work and for the law

The most consequential misreading of safe is the one that hides in plain sight in every workplace, and it has a particularly clear shape in Australian law.

Our everyday speech treats ‘is it safe?’ as a question with a yes-or-no answer, and treats the law as the body that supplies the yes but it does no such thing. The model Work Health and Safety Act, enacted in substance across most Australian jurisdictions, does not impose a duty to make work safe in any absolute sense. The primary duty of care requires a person conducting a business or undertaking to ensure the health and safety of workers and others, but only and always ‘so far as is reasonably practicable’ (Work Health and Safety Act 2011, s 19). The qualifier is not decorative, what is reasonably practicable is defined in the Act as that which is reasonably able to be done to ensure health and safety, weighing up the likelihood of the hazard occurring, the degree of harm, what is known about the hazard and the ways of eliminating or minimising it, the availability and suitability of those ways, and the cost, including whether the cost is grossly disproportionate to the risk (Work Health and Safety Act 2011, s 18). The associated risk-management duty requires that risks be eliminated so far as is reasonably practicable, and where elimination is not possible, minimised so far as is reasonably practicable (Work Health and Safety Act 2011, s 17). Victoria, which retains its own occupational health and safety statute, frames the equivalent employer duty in similar terms, requiring the provision of a working environment that is safe and without risks to health, again qualified by reasonable practicability.

What the statute is doing? Even where the word safe appears in the legislation, it is tethered immediately to reasonable practicability. There is no freestanding legal command to achieve absolute safety, because the lawmakers understood what the philosophers and risk scientists have been saying all along that absolute safety is not a state any continuing activity can reach.

The legal standard is not safe. It is the discharge of a qualified duty to eliminate or minimise risk so far as is reasonably practicable. When a worker, a manager or a director hears ‘we have a duty to keep the workplace safe’ and pictures a guarantee of no harm, they have misread the duty. The duty is to do what is reasonable, to weigh and to act, and to be able to show the weighing. A workplace can fully discharge its legal duty and still be the site of an injury, because the law asks for reasonable diligence against risk, not for the impossible elimination of it.

This is where the philosophical argument earns its keep for the practitioner, because the individual perspectives that Raheemy et al. (2025) found scattered across the profession are not abstractions at all. They are the people in the toolbox talk, the planning meeting and the boardroom, and each of them means something different by the word.

To the worker on the tools, safe most often means felt protection and a confident expectation of going home unharmed. To the supervisor, it tends to mean that the controls are in place and functioning now, a present-tense condition. To the manager, it usually means that the residual risk has been brought to a level they judge acceptable, a relative rather than an absolute claim. To the executive or director, it frequently means an outcome, read backward from incident statistics and assurance dashboards. To the engineer or designer, it means the presence of controls and the capacity of the system to fail without catastrophe, an idea the newer schools of thought have pushed hard (Conklin 2019; Dekker 2014; Provan et al. 2020). To the regulator and the lawyer, it means whether the qualified statutory duty has been met. Those six people are using one word to ask six different questions. When they all ask ‘is it safe?’ and each hears their own meaning in the answer, the shared word produces an illusion of shared understanding that is more dangerous than open disagreement would be, because no one realises the conversation has potentially failed.

The remedy is not a better definition imposed from above but it is a discipline of specification, applied wherever the stakes are real.

The author is hoping that a practitioner who has absorbed the argument of this paper will treat the bare question ‘is it safe?’ as incomplete, and will reflexively unpack it into the questions that actually carry meaning. Safe for whom? which of the people exposed are we talking about, and have we asked the most exposed rather than the most senior? Safe from what: which specific hazards, named, rather than a general aura of safeness? Safe to what standard: acceptable by whose judgement, and shown how? Safe over what horizon: will be, is, or was, and are we inferring a sound process from a lucky outcome? Each of these replaces a comforting adjective with an answerable question, and none takes more than a moment to ask.

Three habits follow this, first, prefer the comparative. ‘This method is safer than that one, for these reasons’ is a claim you can defend; ‘this method is safe’ is a claim that hides a threshold. Second, separate the felt from the controlled. Treat a workforce that feels safe and a workplace that is well controlled as two findings to be checked against each other, not one finding stated twice, because the gap between them is where both complacency and resistance live (Slovic 1987). Third, distrust the past tense, Treat a run of incident-free time as the beginning of an inquiry into whether the work is robust, not as proof that it is, and remember that every uneventful repetition of a degraded practice is exactly what drift feels like from the inside (Rasmussen 1997; Vaughan 1996; Snook 2000; Dekker 2011).

Conclusion

Safe is the word the work safety profession often leans on hardest and questions the least (unfortunately). The plainness of the word can be highly misleading. The word names a judgement rather than a measurement, a value rather than a fact, and a relative standard rather than an absolute one. It runs together what will be safe, what is safe and what was safe. It treats a feeling of safety as if it were the state itself. In law it stands in for a duty that was never meant as a guarantee and  in the hands of government it can lift a highly contestable judgement out of the ordinary people’s debate. Between an employer and a workforce it can be turned either way, to halt the work or to compel it to move forward rapidly. The word is misconstrued not because people are careless but because it fundamentally invites many errors. “Safe” looks like a description of the world, and it is really a verdict we pass on the world, with all the contestability that any verdict carries.

The honest response for us is not to retire the word at all but we need to use it with its hi vis clothes and PPE on. Safer, not safe. For the people doing the work at that time and at that place, not for everyone in general. From these named hazards, not from danger entirely. Acceptable by this standard, and shown to be, not acceptable full stop. Robust across the conditions the work might meet, not merely lucky in the one it met.

A work safety profession willing to speak this way would lose a little of the reassuring certainty the bare word provides. In its place it would gain something far more valuable such as claims that can be examined, decisions that can be defended, and a clearer view of the gap between work as imagined and work as done. That’s the gap is where people are still being hurt.

Raheemy and colleagues (2025) end their study by hoping it will help safety mature as a science, and by noting that, unlike astronomy, the safety community has no single body to govern its definitions. That may be so, but the absence of a governing body does not excuse the individual work safety practitioner, Governments, Unions and Employers from governing their own use of the word.

That work is done one sentence at a time, whenever someone about to say safe, we need to question what they really mean instead.

References