Unsafe acts and unsafe conditions: what they are, where the terms come from, and what to do about them

What the two terms mean, and how to tell them apart on the floor

An unsafe condition is something about the place or the equipment that could hurt someone whether or not anyone does anything wrong. It is there on the night shift when nobody is near it. An unsafe act is something a person does, or leaves undone, that raises the chance or the severity of harm. The act needs a person; the condition does not.

The test on a walk is simple. Take the person out of the picture. If the hazard is still there, you are looking at a condition. If the hazard only exists while someone does the task that way, you are looking at an act. Many reports are both: a guard that has been removed is a condition, and the operator reaching into the opening to clear a jam is an act. Record both. Then ask the question this guide keeps coming back to: what made that act the sensible thing to do at the time?

The table shows the usual pattern. The left-hand column is what a supervisor sees. The right-hand columns are what is usually behind it and the kind of fix that holds, drawn from the hierarchy of controls covered later on this page.

Why 'unsafe act' is a dangerous place to stop

Writing unsafe act on a report feels like a finding. It is closer to a note of where the investigation gave up. Four strands of safety thinking since Heinrich explain why.

Human failure has types, and most of them are not carelessness

The UK Health and Safety Executive's guidance HSG48, Reducing error and influencing behaviour, splits human failure into errors and violations. Errors are unintended: slips (the action goes wrong, like operating the wrong switch), lapses (a step is forgotten) and mistakes (the plan itself was wrong, often from a training gap or a rule applied where it did not fit). Slips and lapses happen to experienced, well-trained people doing familiar tasks, and HSE's point is that training does not remove them; design does. Violations are deliberate, and HSE sorts them too: routine (the shortcut has become how the group works), situational (the rule cannot be followed with the time, staff or equipment available) and exceptional (a rule broken while trying to solve a new problem). HSE lists the design features that breed violations: awkward postures, slow controls, hard-to-reach positions, uncomfortable PPE, procedures that are out of date. Each type calls for a different fix, and almost none of them is about the character of the person.

Active failures and latent conditions

James Reason, in Human Error (1990) and Managing the Risks of Organizational Accidents (1997), separated active failures, the unsafe acts of people at the sharp end whose effects show at once, from latent conditions, the weaknesses that sit in a system for months: a poor design, a maintenance backlog, an understaffed shift, a procedure nobody can follow at line speed. His Swiss cheese picture shows an organisation's defences as slices with holes that move; harm gets through when holes in several slices line up. The act is the last hole. Reason also contrasted the person approach, which treats errors as the product of carelessness and answers with warnings, retraining and blame, with the system approach, which expects people to err and builds defences around them. He was clear that a just culture is not a no-blame culture. It is one that can tell the difference.

Why it made sense at the time

Sidney Dekker's The Field Guide to Understanding 'Human Error' (third edition, 2014) puts the investigator's job in one question: why did it make sense for this person to do this, here, now? People act on what they know, what they are trying to achieve and what has their attention in the moment, which Dekker and others call local rationality. Hindsight hands the investigator knowledge the operator did not have. If reaching into the machine made sense to a competent operator on Tuesday, it will make sense to another one on Friday, and a warning does not change that. Finding out what made it sensible is where the fix is.

Just culture: three kinds of behaviour, three responses

A just culture, in the form David Marx and others have taught to healthcare and aviation, sorts what a person did into three kinds. Human error, an unintended slip or lapse: support the person and fix the system that let it through. At-risk behaviour, a choice whose risk the person did not see or thought was justified, usually because the group has drifted there: coach, and remove whatever made the drift pay. Reckless behaviour, a conscious disregard of a substantial and unjustified risk: this is the rare case that warrants discipline. Most of what gets written up as an unsafe act is one of the first two.

Blame starves the base of the pyramid

The practical cost of blame is silence. People who expect to be written up after a near miss stop reporting near misses, and the plant loses the cheapest signal it had. Our post on why hazards go unreported lists the barriers; fear of repercussions, production pressure and the sense that nothing will change head the list. A falling report count after a disciplinary case is not an improvement. It is the base of the pyramid going dark.

A working method for the plant

Six steps, in the order a report moves through them. They work for a safety observation card, a near miss report, a tag raised from the floor, or something a supervisor sees on a walk.

1. 1

   Record every report as act, condition or both, then ask the system question

   Keep the two words; they are useful. Add a third field to the report, card or tag: what made this the sensible thing to do? The reporter rarely knows the whole answer, but the question tells everyone that the report is about the work, not the worker. On a walk, when you see an act, look for the condition behind it before you say anything, and ask the person rather than tell them.

2. 2

   Triage with one rule

   If the hazard could kill or maim and is present now, stop the task and put an interim control in place today: a barrier, a lock, a sign and a named person who owns it until the real fix is in. Everything else gets an owner and a due date within the day and goes on the board. Severity first, then how often people are exposed, then how easy the fix is. The risk assessment matrix is enough for this and does not need a scoring debate.

3. 3

   Write a corrective action that survives

   Four things make an action real: a named owner, a due date, the level of the hierarchy it sits on, and how it will be verified. Remind operators about the guard has no level and no verification. Fit an interlocked hinged guard to transfer 3, owner maintenance lead, due the 14th, engineering control, verified by a 30-day check that the guard stays closed and a count of jams, can be checked by anyone. The corrective action form has a place for each, and the job safety analysis template is where the new standard for the task gets written.

4. 4

   Verify before you close

   Closing means the control is in, it works, and it has not created a new hazard. OSHA's guidance puts it as confirming that controls were implemented as planned and are working. Go back to the spot, watch the task once, ask the operator. If the fix was administrative, set a date to check whether it has held, because those are the fixes that fade.

5. 5

   Close the loop in public

   Tell the person who reported it what was done, and show the fix at the next tier meeting or toolbox talk. Reports rise when people see them turn into changes. Share the fix with the other lines and plants that have the same machine, which is what yokoten means. A serious case gets a full incident investigation and the root cause analysis tools, with the act, the condition and the system causes all in the report.

6. 6

   Track leading indicators, not only injuries

   Count the things that come before harm: observations and hazard reports made, near misses reported, the share of reports that named a condition or a system cause rather than only an act, time from report to close, actions overdue, and the share of closed actions at the top three levels of the hierarchy. A rising report count with a falling time to close is a plant that is learning. Set the baseline from your own numbers; no outside figure says what good looks like for your plant.

Common mistakes

• Quoting 88% as a fact. It is a 1931 claim from records that no longer exist, and the safety professional who reviewed Heinrich's editions most closely asks the profession to stop repeating it.
• Counting acts and conditions as if the split meant something. The ratio in your reports says more about who writes them, and what they are allowed to write, than about your plant.
• Closing a report with retrain, remind or warn. Those are the weakest controls available, and the only ones that need repeating.
• Disciplining after a near miss. The report count falls, the hazards stay, and the next event arrives without warning.
• Treating observation cards as the programme. Cards find drift; only a fix on the hierarchy removes it.
• Fixing the act in the report and leaving the condition on the floor. If the guard is still slow to open, the next operator will tie it back too.
• Reading a falling near miss count as success. Check first whether reporting got harder, or riskier, or stopped being answered.

Definitions and sources

• Heinrich's triangle (300-29-1) and causation split (88-10-2), the 12,000 insurance claim cases and 63,000 plant-owner cases, the changes in wording between editions, and the instruction to stop at the first proximate and most easily prevented cause: Fred A. Manuele, Reviewing Heinrich: Dislodging Two Myths From the Practice of Safety, Professional Safety, October 2011, which quotes and cites each edition of H. W. Heinrich, Industrial Accident Prevention (first edition 1931, fourth edition 1959). Heinrich's book itself was not read for this page.
• Bird's ratio (1 : 10 : 30 : 600) from the accident reports of 297 companies, 1969, and the note that serious injuries and fatalities often have different precursors from minor injuries: as stated on our safety pyramid poster page, which gives its sources.
• Human failure types (slips, lapses, mistakes; routine, situational and exceptional violations), the design features that increase violations, and the warning against blaming the individual directly involved: Health and Safety Executive, Reducing error and influencing behaviour, HSG48, second edition 1999, and the HSE human factors pages on managing human failures.
• Active failures and latent conditions, the Swiss cheese model, and the person approach against the system approach: James Reason, Human Error (Cambridge University Press, 1990); Managing the Risks of Organizational Accidents (Ashgate, 1997), where the just culture idea is set out; and Human error: models and management, BMJ 2000, volume 320, pages 768 to 770.
• Local rationality and the question of why it made sense at the time: Sidney Dekker, The Field Guide to Understanding 'Human Error', third edition (Ashgate, 2014).
• The three behaviours of a just culture (human error, at-risk behaviour, reckless behaviour) and the responses to each: David Marx's just culture materials as presented in his conference talks and the Outcome Engenuity model, paraphrased.
• Hierarchy of controls: NIOSH, Hierarchy of Controls (cdc.gov/niosh), including the point that the top of the hierarchy works without relying on worker behaviour and the Prevention through Design note; OSHA, Recommended Practices for Safety and Health Programs, Hazard Prevention and Control (interim controls, follow-up to confirm controls work); ISO 45001:2018 clause 8.1.2, Eliminating hazards and reducing OH&S risks (clause title only; the standard's text is not reproduced).
• What was not used: no industry rate for unsafe acts, unsafe conditions, near misses or injuries appears on this page, because none could be traced to a study that applies to your plant. The only numbers are the attributed historical ratios above and the made-up figures in the worked example.