Robot safety
Why robot-death counts are overstated: the keyword trap
A worker collapsed and died in a robotic weld area. A keyword search of fatality data for 'robot' returns his death, but the cause was asthma, natural causes, and the robot had nothing to do with it.
That case is why raw robot-fatality counts, the kind built by searching incident databases for the word 'robot,' overstate the toll. The keyword catches every death that happened near a robot, not only the ones a robot caused.
This page traces one such record to its source and shows why a robot-death count needs reading, not just counting.
Data covers How keyword-based robot-fatality counts overstate, via an OSHA case record. Published 2026-07-20. Last reviewed 2026-07-20. Last updated 2026-07-20. Edited by Mike Ramsey / Reliable Media.
The figures and where they come from
Each figure is rated for how safely you can cite it today. Ratings judge current usability, not whether a number was ever correct.
| Figure | What it is | Source | Citation Confidence | Notes |
|---|---|---|---|---|
| asthma, natural causes | Recorded cause of death | [A] | High | The OSHA report states the worker died due to asthma and the cause of death was natural. Not a robot fatality, despite surfacing in a robot-keyword search. |
| in a robotic weld area | Where it happened | [A] | High | The death occurred in a robotic weld area, which is why a keyword search for 'robot' returns it. Proximity, not causation. |
Why the numbers disagree
A keyword count answers the wrong question. Searching a fatality database for 'robot' returns every record that mentions the word, which includes deaths that merely happened in a robotic area, not deaths a robot caused. This OSHA record is exactly that: a worker who died of asthma, from natural causes, in a robotic weld area. Counting it as a robot fatality is a category error.
That is why the aggregate robot-death figures that circulate should be treated with care. Some are built by keyword-searching public incident data, which inflates the total with proximity deaths, and others trace to a paywalled paper that cannot be verified. Neither method produces a clean, citable count.
The fix is reading, not counting. Each record has to be checked for whether the robot caused the death, which is why the individually-documented cases, where the mechanism is clear, are more useful than any keyword-derived total. A number is only as good as the records behind it, read one by one.
How to cite these figures
Do not quote a robot-fatality count built by keyword-searching incident data. It includes deaths that happened near a robot but were not caused by one.
If you cite robot fatalities, use individually-documented cases where the mechanism is established, not an aggregate keyword total.
State the caveat plainly: proximity to a robot in an incident record is not causation, and raw counts overstate the toll.
Where people go wrong
Treating a keyword search for 'robot' as a robot-fatality count. It catches proximity deaths, like a natural-causes death in a robotic weld area.
Assuming public incident databases yield a clean robot-death total. They require case-by-case reading to separate cause from proximity.
Repeating an aggregate figure without knowing how it was built. Keyword-derived counts overstate; the paywalled academic figure cannot be verified.
How we checked
We traced one over-counted record to its OSHA accident report and confirmed the cause, asthma and natural causes, and the setting, a robotic weld area, in its text.
We use this single record to demonstrate a counting method's flaw, not to make a claim about frequency. The point is that a keyword search conflates proximity with causation.
We connect it to the broader problem covered elsewhere on the site: the aggregate robot-death figures either come from keyword counts that overstate or from a paywalled paper that cannot be verified, so no clean count exists.
Full source list
Primary sources, with live links. Every figure above traces to one of these.
- [A]U.S. Occupational Safety and Health Administration2023
OSHA Accident Report 155569.015, "Employee dies of natural causes" (worker in a robotic weld area, cause asthma)
https://www.osha.gov/ords/imis/accidentsearch.accident_detail?id=155569.015
Common questions
- Are robot-fatality counts accurate?
- Counts built by searching incident data for the word 'robot' overstate the toll, because they include deaths that happened near a robot but were not caused by one, like a worker who died of asthma from natural causes in a robotic weld area.
- How does a non-robot death end up in a robot count?
- By keyword. A search for 'robot' returns every record that mentions the word, including a death that merely occurred in a robotic area. Proximity is not causation.
- So how many people do robots kill?
- There is no clean, citable count. Keyword-derived totals overstate by including proximity deaths, and the most-cited academic figure is behind a paywall and cannot be verified. Documented individual cases are more reliable than any aggregate.
- What is the right way to count?
- Case by case. Each record has to be read to confirm the robot caused the death, which is why individually-documented cases beat a keyword-derived total.
More data, traced to source
- Robot workplace deaths: what the public record does and doesn't show
The figure of about 41 US robot-related deaths is quoted widely, but it cannot be verified through any free primary source, and the same agency has published two different totals. This page traces what is and is not documentable about robot workplace fatalities.
- What robot fatality case files actually record
The individual robot-fatality investigations that do exist point to one failure again and again: the robot was live when it should have been locked out. What the OSHA and NIOSH-family case files record.
- How workers die on the job in the EU: struck, crushed, machines
More than a quarter of EU workplace deaths, 27.1%, come from losing control of a machine or equipment. Another 21.4% are struck by a moving object and 13.8% trapped or crushed. The contact modes behind the machine and robot hazard.