A Robot Injuring a Worker During Autonomous Operation: What Evidence Could Prove—or Defeat—Liability?

The evidence that decides a robot injury case is almost always the machine's own record: controller logs, safety-circuit history, and the maintenance and lockout paperwork from the minutes before the injury. Those records show whether the robot did something it was not designed to do, or whether a person was inside its working envelope while its stored energy was still live — and that distinction usually decides who pays.

"Autonomous operation" here means the robot was running its program under its own control, not being hand-jogged by an operator. That matters because the classic robot injury is not a machine going rogue. It is a machine doing exactly what it was told, resuming a cycle at the moment a worker had stepped into the cell.

The paper trail that has already decided these cases
What to preserve in the first week
The standards that give an expert something to measure against
What defeats the claim
Europe is rewriting who has to produce the evidence
Practical steps for an injured worker or a family
Frequently Asked Questions

The paper trail that has already decided these cases

The most-cited robot fatality in American enforcement is instructive about what evidence carries weight. Regina Elsea, 20, was crushed on June 18, 2016 at Ajin USA when a robot restarted as she and three coworkers entered a robotic cell to clear a sensor fault without locking out stored energy. An administrative law judge affirmed the resulting citations and the company paid over $1.3 million, according to OSHA's account of the case. Nothing in that outcome turned on a software bug.

It turned on the energy-control procedure — whether one existed, whether it was written for that cell, and whether anyone followed it. OSHA's original case included instance-by-instance willful violations for failing to use energy-control procedures during servicing, and placed the company in the Severe Violator Enforcement Program, per the Department of Labor's 2016 release. The same incident also produced a federal criminal resolution: the statutory maximum $500,000 fine, $1,000,000 in restitution to Elsea's estate, and three years' probation, documented by the Justice Department. That record exists independently of any civil suit — which is the practical point for a reader. Enforcement files, inspection narratives, and criminal dockets are public documents that a claimant's lawyer can obtain without waiting for discovery from the employer.

What to preserve in the first week

Robot cells overwrite their own evidence. Controller error stacks roll over, cycle counters reset on restart, and video systems recycle storage on a fixed schedule.

A preservation letter sent early is often worth more than any expert report filed a year later. The items that typically decide fault: The single most revealing document is often the maintenance history. A cell that jams predictably, and that workers have learned to clear quickly without a full shutdown, builds the employer's knowledge of the hazard into its own records.

Robot controller logs — program name, cycle position, fault codes, and the timestamp of the restart command.
Safety PLC or safety relay history — which interlock opened, when, and whether it was bypassed or muted.
Lockout/tagout records — the written procedure for that specific cell, the lock assignment, and training sign-offs.
The light curtain, area scanner, or door interlock configuration file, including any muting zones.
Maintenance tickets for the preceding weeks, especially repeat faults that trained workers to reach in rather than shut down.

The standards that give an expert something to measure against

There is no OSHA standard written specifically for robots. Cases are built on the General Duty Clause 5(a)(1), machine guarding at 29 CFR 1910.212, and lockout/tagout at 29 CFR 1910.147 — and OSHA's own 1987 robotics directive, STD 01-12-002, was formally cancelled on March 15, 2023, removing the agency's dedicated robotics guidance document. Consensus standards fill that space. ISO 10218-1 and -2 were republished in February 2025, the first major revision since 2011.

The old blanket requirement that safety-related control functions hit Performance Level d / Category 3 is replaced by per-function performance levels, chosen from a table or from a documented risk assessment, and the collaborative limits from ISO/TS 15066 are absorbed into the standard, per ISO's record of the revision. For a plaintiff, that shift cuts both ways: the documented risk assessment becomes discoverable, and a missing one is conspicuous. Where a collaborative robot touched a person, there is an actual number to test against. ISO/TS 15066's Annex A sets maximum contact force and pressure for 29 body regions, derived from a 100-subject pain-onset study run at the University of Mainz for the ISO committee, with clamping limits set well below transient-impact limits. A biomechanical expert can compare measured cell forces against the limit for the body region struck — a rare instance of a bright line in an injury case.

What defeats the claim

Three defenses recur, and each is beaten or lost on the same records described above. The first is that the robot performed to specification and the worker entered a guarded zone. If the interlocks were functional and logged, and the energy-control procedure was written and trained, the manufacturer walks away and the case narrows to the employer — where workers' compensation is generally the exclusive remedy against a direct employer, leaving a third party as the only route to full damages. The second is that someone bypassed the safeguards. Defeated interlocks, taped light curtains, and disabled area scanners are visible in the safety circuit history and in the hardware itself. That finding can shift a case from product liability toward employer conduct — or toward an integrator who shipped the cell with an easy defeat.

The third is that no defect can be proven. This is the structural problem with robot litigation: the injury population is small. A peer-reviewed census-based study counted 41 robot-related worker fatalities in the United States between 1992 and 2017, published in the American Journal of Industrial Medicine. A separate review of OSHA Severe Injury Reports found 77 robot-related accidents from 2015 through 2022 — 54 stationary-robot accidents causing 66 injuries, mostly finger amputations and head or torso fractures, and 23 mobile-robot accidents causing 27 injuries, mostly leg and foot fractures, reported in *Applied Ergonomics*. Those numbers are too small to support the industry-wide failure-rate statistics that anchor a classic design-defect argument. Proof has to come from this cell, this program, and this incident.

Europe is rewriting who has to produce the evidence

The same case brought in the EU will soon run on different rules. Directive 2024/2853 treats standalone software as a "product," allows courts to order a defendant to disclose relevant evidence, and directs courts to presume the product defective when the defendant fails to disclose, or when proof is excessively difficult because of technical complexity. Defendants may rebut the presumption.

Member States must transpose it by December 9, 2026. Alongside it, the EU AI Act requires high-risk AI systems to automatically record events across their lifetime under Article 12, and requires providers to retain those automatically generated logs for at least six months under Article 19. The practical effect is that the machine's own record becomes the decisive evidence — and its absence is itself a compliance failure rather than a gap the injured person has to work around.

Practical steps for an injured worker or a family

If a robot injury happened during normal automatic operation, the first question a lawyer will ask is not what the software did. It is whether the energy was locked out before anyone crossed the fence line — and the answer is written down somewhere before anyone thinks to look for it.

File the workers' compensation claim immediately; it runs on its own deadline, separate from any lawsuit.
Ask, in writing, that the employer preserve controller logs, safety circuit history, and video. Name the cell and the date.
Request the OSHA inspection file once an inspection opens. Citations, narratives, and photographs become obtainable records.
Identify every non-employer in the chain: robot manufacturer, cell integrator, guarding supplier, the contractor servicing the machine, and any staffing agency involved.
Photograph the cell before it is rebuilt. Guarding is routinely modified within days of a serious injury, and the "after" state is not evidence of what existed.

Frequently Asked Questions

Can I sue my employer if a robot injured me at work?

In most cases workers' compensation is the exclusive remedy against a direct employer, so the fuller recovery usually comes from a third party — the robot manufacturer, the cell integrator, the guarding supplier, or a servicing contractor. Identify those parties early, because their evidence sits outside the employer's control.

Does an OSHA citation prove my case?

It does not decide a civil claim by itself, but it produces documents, measurements, and findings you would otherwise pay an expert to develop. In the Ajin USA matter, citations were affirmed by an administrative law judge and the company paid over $1.3 million, creating a public record independent of any civil suit.

What if the robot's logs were not kept?

In the United States, a missing log is argued as spoliation and its consequences depend on the court. Under EU Directive 2024/2853, courts may order disclosure and presume defectiveness when a defendant fails to disclose, subject to rebuttal — a materially different starting position.

Are mobile robots treated differently from fixed arms?

The injury pattern differs. The *Applied Ergonomics* review of 2015–2022 OSHA Severe Injury Reports found stationary robots mainly caused finger amputations and head or torso fractures, while mobile robots mainly caused leg and foot fractures. Mobile units also raise navigation, mapping, and floor-marking evidence that fixed cells do not.