American factories have never been more automated, and body parts caught in running machinery remain the single most common cause of severe workplace injury in the country. Between 2015 and 2021, employers reported 12,930 such incidents to federal OSHA, 18% of all severe injuries logged in that period. Manufacturing accounted for 8,904 of them, or 69%. Over the same seven years, US robot density climbed steadily, reaching 307 industrial robots per 10,000 manufacturing employees by the International Federation of Robotics’ 2025 count.
Those two trends are supposed to move in opposite directions. They don’t, and the reason has less to do with robots than with when humans put their hands inside them.
Automation removes the operator from routine contact with the machine. It does not remove the jam, the misfeed, the sensor that needs wiping, or the changeover. Those interventions are unplanned, unscripted, and frequently performed on equipment that is still energised. The injury data reflects that shift with uncomfortable clarity.
What struck me working through the severe injury reports is how tightly the incidents cluster into the same few minutes of the day. Not the production run. The clear-the-fault, reach-in, get-it-moving-again minutes that appear on no process diagram. Plants measure uptime and cycle time to four decimal places, and almost none of them measure how often somebody puts a hand past a guard to keep the line running. That’s the number I’d want first if I were writing the safety case.
What the Severe Injury Data Actually Shows
Since January 2015, US employers have been required to report any work-related amputation, inpatient hospitalization, or loss of an eye to OSHA within 24 hours. The resulting dataset is the closest thing the country has to a real-time register of catastrophic workplace injury.
OSHA’s seven-year lookback, covering 2015 through 2021, recorded 70,206 severe injury reports. Of those, 56,696 were inpatient hospitalizations and 18,559 were amputations. A further detail buried in the methodology matters more than the headline: 5,049 of those amputations were severe enough to also require hospitalization. The agency estimates the report covers roughly half of the US workforce, because private-sector employees in state plan states and most state and municipal workers fall outside federal jurisdiction.
Double the figure to approximate national coverage and the arithmetic becomes difficult to ignore. Somewhere around 5,000 American workers lose part of a limb at work every year.
The rate column tells a different story from the count column. Mining posts the highest rate per 100,000 full-time equivalent workers at 45.3, well above manufacturing’s 26.5, but generates a fraction of the total volume. Manufacturing dominates the absolute numbers because that is where the machines and the workers both are.
Within the caught-in-machinery category, plastics product manufacturing was the single highest-reporting industry at 701 incidents, about 5% of the category.
The Maintenance Window Is Where Injuries Cluster
OSHA publishes narrative summaries alongside the counts, and the pattern in them is consistent enough to function as a diagnosis.
One worker’s glove became entangled between a conveyor belt and the belt tensioning roll on an unguarded conveyor, taking a finger. Another partially lost a finger while unjamming a de-tailer at a packing house. A third was removing pizza from a conveyor line when a gloved finger caught between the belt and a roller, producing a fracture, a fingertip amputation, and a nail avulsion. A fourth was working on a wood waste belt when an arm was drawn into the power transmission drive and amputated at the bicep.
Not one of those describes a worker performing the machine’s designed task. Every one describes a worker fixing, clearing, or reaching into it.
Research on robot-related fatalities points the same direction with a sharper number. A CDC and NIOSH analysis found that 58.5% of robot-related workplace fatalities occurred during maintenance tasks, including unjamming equipment and cleaning sensors, rather than during normal automated operation. The machine was not doing something unexpected. A person was doing something the risk assessment treated as an edge case.
The control for this is well established and poorly applied. OSHA’s Control of Hazardous Energy standard, 29 CFR 1910.147, governs the isolation of energy before anyone reaches into equipment. In the FY2025 enforcement figures, lockout/tagout ranked fourth among all cited standards with 2,177 citations, moving up a position from the previous year. It has never left the top ten.
What the Guarding Rule Was Written Around
The governing regulation is short enough to quote. Under 29 CFR 1910.212(a)(1), one or more methods of machine guarding must protect workers “from hazards such as those created by point of operation, ingoing nip points, rotating parts, flying chips and sparks.”
Those four categories map onto four distinct mechanisms of injury. The point of operation is where the machine does its work, the press that closes, the blade that cuts. An ingoing nip point is the line where two rotating parts draw together, which is why conveyor belts and their tensioning rollers appear so often in the injury narratives; a glove entering a nip is pulled in at machine speed, and the hand follows. Rotating parts include shafts, couplings and power transmission drives that continue turning long after the productive part of the cycle has stopped. Flying chips and sparks are the outlier, causing eye and burn injuries rather than amputations.
The next subsection is where the gap opens. Section 1910.212(a)(3)(ii) requires that point of operation guarding be designed “to prevent the operator from having any part of his body in the danger zone during the operating cycle.”
During the operating cycle. That phrase describes the exact window in which almost none of the reported amputations occurred.
This isn’t a drafting error so much as a rule written for a different generation of equipment, one where the operator stood at the machine and fed it by hand throughout the shift. Guarding the operating cycle was the whole problem in 1971. In an automated cell, the operating cycle is the safest period of the day, because nobody is anywhere near the machine. Risk concentrates in the interruptions, and those are governed by a separate standard that a separate group of workers is trained on, often to a lower level of rigour.
Why Machine Guarding Citations Understate the Problem
Machine guarding, 29 CFR 1910.212, ranked tenth in OSHA’s FY2025 most-cited list with 1,239 citations. Read casually, that looks like a hazard being brought under control. Read against the injury data, it looks like something else.
Citation counts measure inspection activity, not hazard prevalence. Fall protection topped the FY2025 list with 5,914 citations largely because construction sites are visible, numerous, and inspected often. A guarded point of operation inside a sealed automated cell is not visible from the street, and the violation is frequently discovered only after the injury that prompts the inspection.
OSHA’s own estimate of the underlying burden is far larger than any citation total suggests. The agency puts the toll among workers who operate and maintain machinery at roughly 18,000 amputations, lacerations, crushing injuries and abrasions plus more than 800 deaths per year, with about half of all amputations occurring in the manufacturing sector.
Enforcement has been targeted accordingly. OSHA published CPL 03-00-022, the National Emphasis Program on Amputations in Manufacturing Industries, in December 2019, superseding the 2015 version. It covers 75 NAICS codes spanning meat and poultry processing, bakeries, wood, plastics, and metal fabrication, and it directs inspectors to use employers’ own amputation reports as targeting data. A November 2024 memorandum extended the program.
An emphasis program is a reasonable response to a persistent hazard. It isn’t evidence that the hazard is shrinking.
What Changed in the Robot Safety Standards in 2025
The international standards community reached a similar conclusion and rewrote the rulebook around it. ISO 10218-1 and ISO 10218-2 were both republished in 2025, the first substantial revision of the industrial robot safety standard since 2011.
The most consequential change is terminological, which sounds like a lawyer’s quibble until you trace what it fixes. The revised standard drops “collaborative robot” in favour of “collaborative application,” on the reasoning that only an actual deployment can be designed, tested, and verified as safe. A robot arm sitting on a pallet has no safety rating. The same arm, running a specific payload at a specific speed next to a specific worker doing a specific task, does.
That distinction corrects a category error the market spent a decade making. Manufacturers sold cobots as inherently safe machines, buyers installed them without fences, and the risk assessment that was supposed to accompany the deployment often didn’t happen.
Most of the technical content of ISO/TS 15066 has now been folded into ISO 10218-2, including the force and pressure limits it defined across 29 separate body regions. The revision also clarifies functional safety requirements and, for the first time, adds cybersecurity requirements to a robot safety standard, an acknowledgement that a compromised controller is a mechanical hazard rather than only an IT problem.
None of which is law in the United States. OSHA states plainly that there are no specific OSHA standards for the robotics industry, and regulates robot cells through general industry rules instead, principally machine guarding under Subpart O, control of hazardous energy under 1910.147, electrical requirements under Subpart S, and personal protective equipment under Subpart I.
The consensus standard fills the gap, with a lag worth noting. OSHA’s robotics page points to ANSI/RIA R15.06-2012, described as the US national adoption of ISO 10218-1,2:2011. That means the American reference document is anchored to a 2011 international standard, while the international standard itself moved on in 2025. Any plant buying a cell today is working from guidance written before collaborative applications were common, before force and pressure limits were formalised in ISO/TS 15066, and well before anyone considered controller cybersecurity a safety matter.
The Cost That Lands on the Worker
An employer’s reporting obligation for an amputation closes in 24 hours. The worker’s version of the same event runs for the rest of their working life, and the accounting is asymmetric in ways that rarely appear in safety statistics.
Prosthetic devices are consumable. Clinics plan on full replacement every three to five years, with annual maintenance and supplies adding 5% to 10% of the device cost. A myoelectric arm runs $20,000 to $50,000, and multi-articulating bionic hands start around $50,000, which means the lifetime figure for a hand lost at 30 is a different order of magnitude from the invoice for the first device. Anyone pricing the consequences of a machine guarding failure should start with what a bionic arm actually costs across a replacement schedule rather than with a single quote.
Insurers already price amputation as the most expensive injury on the board. NCCI data published through the National Safety Council’s Injury Facts puts the average lost-time workers’ compensation claim for an amputation at $125,058 for claims occurring in 2022 and 2023. The next most costly categories trail well behind, with other trauma at $68,231, fracture, crush or dislocation at $66,467, and burns at $64,019. Against an average of $47,316 across all lost-time claims, an amputation runs about 2.6 times the norm.
Set that $125,058 next to the device figures above and the shortfall is the whole story. One myoelectric arm can consume a third of it. A replacement every three to five years over a 30-year working life exhausts it several times over, before counting therapy, socket refits, or the second device most people need for work or water.
Workers’ compensation covers medical treatment and a portion of lost wages, and in exchange it generally bars the employee from suing the employer. It doesn’t bar claims against third parties, which is why the machine builder, the integrator who installed the cell, and the maintenance contractor all become relevant once the facts are established. Preservation of the equipment in its post-incident state matters enormously to that analysis, and it is routinely lost when a line is cleaned and restarted the same shift.
Where the Risk Concentrates Next
US robot density stood at 307 units per 10,000 manufacturing employees in the IFR’s World Robotics 2025 report, placing the country eighth globally and two positions higher than the previous year. Installations reached 38,000 units in 2025, an 11% annual increase.
Against the leaders, the US is mid-table. South Korea runs 1,220 robots per 10,000 workers, Germany 449, Japan 446. China, at 166, is climbing fast from a lower base.
So where does the exposure go next? The implication is straightforward and unwelcome. The US isn’t past its automation build-out. It’s inside it, which means the population of recently installed cells, newly trained maintenance staff, and unfamiliar failure modes is growing rather than stabilising. Every one of those cells will jam. What determines the injury rate is whether clearing that jam is a documented, energy-isolated procedure or an improvisation performed by whoever is nearest.
Frequently Asked Questions
How many workplace amputations happen in the US each year
Employers reported 18,559 amputations to federal OSHA between 2015 and 2021, an average of about 2,650 per year. That figure covers roughly half the US workforce, since state plan states and most public employees fall outside federal jurisdiction, so the national total is considerably higher.
Which industry has the most machine-related amputations
Manufacturing, by a wide margin. It accounted for 8,904 of the 12,930 severe injuries involving body parts caught in machinery from 2015 to 2021, or 69%. Plastics product manufacturing was the highest single reporting industry within that category. OSHA also estimates that about half of all workplace amputations occur in manufacturing.
Do collaborative robots need safety fencing
It depends on the application, which is precisely the point of the 2025 revision to ISO 10218. Safety is a property of the deployment rather than the robot, so a risk assessment of the installed application determines whether fencing, speed limits, force limits or other measures are required. No robot is inherently safe independent of what it’s holding and how fast it’s moving.
Why does lockout tagout matter more in automated plants
Because automation shifts human contact from scheduled operation to unscheduled intervention. NIOSH found that 58.5% of robot-related fatalities happened during maintenance work such as unjamming and sensor cleaning. Lockout/tagout under 29 CFR 1910.147 is the standard governing exactly those moments, and it drew 2,177 citations in FY2025.
Is machine guarding still a common OSHA violation
Yes. Machine guarding under 1910.212 placed tenth in FY2025 with 1,239 citations and has held a top ten position for more than two decades. Manufacturing employers receive the large majority of those citations.
How automated are US factories compared with other countries
The US had 307 industrial robots per 10,000 manufacturing employees in the IFR’s 2025 report, ranking eighth worldwide. South Korea leads at 1,220, followed by Germany at 449 and Japan at 446. US installations grew 11% to 38,000 units in 2025.
The Metric That Would Actually Move
Safety programmes in automated plants tend to measure the wrong thing. Recordable incident rates are lagging indicators by construction, and guard audits check whether a barrier exists rather than how often someone goes around it.
The number that predicts an amputation is the count of unplanned entries into a machine’s danger zone per week, multiplied by the share of those entries performed without energy isolation. Plants that instrument that figure find it is larger than management assumed, usually by a factor that makes the subsequent conversation short. If US robot installations keep growing at 11% a year while that number stays uncounted, the severe injury data will keep saying what it has said since 2015.
