Left Unattended
The rules that govern American pits were written to protect the person in the cab. The trucks running Texas quarries today have no one in them.
Start with the record.
Every shift I ever ran, a person walked the machine before it moved. Tires, hoses, lights, brakes, fluids, guarding. Then that person signed a preshift record, and if a defect showed up, it was on paper, and it was on somebody.
That record is not a formality. It is the load bearing wall of surface mine safety in this country. The rule that requires it, 30 CFR 56.14100, says in plain language that self propelled mobile equipment must be inspected by the equipment operator before it is placed in operation on that shift.
By the equipment operator.
Read that again with a driverless haul truck in mind.
There is no operator in the cab. There is no one to walk it, no one to sign it, no one whose name lands on the line when the truck rolls out. The rule assumes a human being that the machine no longer has.
That is not a scandal. It is a gap. And that gap is not waiting on some distant future to show up. The trucks are already running.
I. It is already running
This is not a forecast. It is a quarry in Texas.
At Heidelberg Materials’ Lake Bridgeport operation, a mixed fleet of Caterpillar and Komatsu haul trucks has been running on Pronto AI’s autonomous system with no driver in the cab. In under eight months it moved more than two million tons of limestone. Pronto calls it North America’s first fully autonomous mixed fleet quarry operation, and Heidelberg has committed to putting more than a hundred self driving trucks across its sites by 2028.
It is also a quarry in Virginia. Luck Stone’s Bull Run plant went live with autonomous Cat 777s in late 2024 and passed two million tons hauled within its first year, on a single shift operation, with no reported safety injuries.
Over a year of driverless production on MSHA regulated ground, at multiple sites, and not one autonomy driven MSHA problem to show for it.
Sit with that, because it cuts both ways.
One way, it is the best argument autonomy has. The technology works. The operators are good. Taking the driver out of the cab took a human out of the path of the most dangerous machine in the pit. Anyone who wants to call autonomy reckless has to explain millions of driverless tons and nobody hurt.
The other way is the one nobody wants to sit with. A clean record is not a resolved rulebook. It is an untested one. Every hard question about who inspects, who is accountable, and who pays stays asleep until the day something goes wrong. No incident means no citation, no lawsuit, no precedent. The framework has not passed.
It has not been called yet.
II. The paradox: the safer it gets, the less the rules apply
Here is the part of this story almost nobody says out loud.
The Mine Act exists to protect miners. Every standard under it, every citation written against it, traces back to a human being who could get hurt. That is the whole architecture. The regulations do not protect trucks. They protect people near trucks.
Now follow that logic into an autonomous pit.
MSHA leadership has indicated in public settings that unless there is a specific hazard to a miner in close proximity to an autonomous vehicle, enforcement personnel would probably not consider the regulations applicable. The rules are focused on miner safety. No miner in the area, no clear enforcement hook.
Think about what that means.
The safer autonomy makes the pit for people, the less regulatory grip exists on the machines. Pull every human out of the haulage circuit and you have not just reduced risk. You have partially removed the machine from the enforcement frame the rules were built on. A two hundred ton truck making its own decisions at forty kilometers an hour, and the regulatory system’s honest answer is that its rules mostly attach when a person wanders close.
I am not saying MSHA is asleep. District offices are engaging, and the agency has said it is open to working with the industry on this equipment. But there is no autonomy standard. There is no rulemaking for one. And the current regulatory direction runs the other way entirely: in July 2025, MSHA published eighteen proposed rules in a single sweep, aimed at modernizing old standards and reducing compliance burden under a deregulatory executive order.
Read that agenda however you want. What it is not, is an autonomy rulebook in progress.
So the question I asked at the top, who writes the rules for the unattended brain, now has a real answer.
Nobody is writing them. Which means the first serious incident will. Through citations stretched over rules from 1985, through discovery fights, through a jury. Unless the industry writes them first.
III. What the rules actually say
Here is the honest state of play. It is not a vacuum. It is a framework built for a manned era, doing its best to stretch over machines that removed the man.
The most relevant recent action is MSHA’s Safety Program for Surface Mobile Equipment rule. Published December 2023, effective January 19, 2024, enforced from July 17, 2024. It lives in 30 CFR at 56.23000, 57.23000, and 77.2100.
It requires every operator to build and maintain a written safety program for surface mobile equipment, developed with input from miners and their representatives. The program has to identify and analyze hazards, reduce risks, maintain procedures and schedules for routine maintenance and non routine repairs, and identify currently available and newly emerging feasible technologies that can improve safety.
That last phrase is the door autonomy walks through. The definition of surface mobile equipment is broad enough to sweep in autonomous and semi autonomous vehicles without ever naming them. So the framework technically covers driverless trucks today.
But notice what kind of rule it is.
Performance based, not prescriptive. It tells you to write a program and manage your hazards. It does not tell you how to inspect a machine that inspects itself, how to validate a perception model, how to handle a software update that changes truck behavior overnight, or what a digital examination record must contain to satisfy an inspector. A good general rule, doing work it was not designed for.
Underneath it, the older standards still govern, and this is where the friction lives.
30 CFR 56.14100 requires that preshift inspection by the equipment operator. On a driverless truck, who is the operator. The control room. A designated person. The maintenance tech who did the last walkaround. The rule does not say, because when it was written the question could not exist.
30 CFR 56.14207 covers parking procedures for unattended equipment. Sounds purpose built for this. It is not. Read it and you find a rule about a machine that is parked. Controls in park, brake set, wheels chocked or turned into the bank on a grade. It governs a truck sitting still with nobody in it. It has nothing to say about a truck moving through the pit with nobody in it, because in 1985 that was not a category of thing. The regulatory vocabulary does not have a word for equipment that is operating and unattended at the same time.
30 CFR 56.18002 requires a competent person to examine each working place every shift. When the working place is partly a fleet of robots, what does that examination look like, and who is competent to perform it.
I have written those records. Preshift, working place, defect logs, the whole paper spine of a compliant operation. I know what a person is doing when they fill one out, and it is not checking a box. It is a trained human making a judgment call about whether a machine is safe to put in front of other humans, and putting their name on that call.
The truck is very good at a lot of things. That judgment, and the accountability attached to it, is the part nobody has cleanly transferred.
Sidebar: Same rule, new machine
Key 30 CFR provision Manned equipment Autonomous system 56.14100 Preshift exam Operator inspects the machine before the shift and signs the record No operator in the cab. Who inspects, who signs, and what counts as the exam is undefined 56.14207 Unattended equipment Governs a parked machine: park, brake, chock A moving driverless truck is neither parked nor attended. The rule has no category for it 56.18002 Working place exam Competent person walks and examines the area each shift The working place now includes robots. Who is competent to examine a mixed autonomous area is unsettled Defect reporting and records Human observes, reports, and records the defect The machine’s data log is the record, and it is proprietary and opaque
IV. This is not arriving in the Pilbara. It is arriving in quarries.
Here is the part I have not seen anyone else write, and it is the part I know from the inside.
Autonomy built its safety record in Tier 1 iron ore. Purpose designed haul roads. Segregated autonomous zones. Fleets of five hundred ton trucks running the same loop for years. Control rooms in Perth staffed around the clock. Engineering departments bigger than most quarries’ entire headcount. The Pilbara record is real, but it was earned inside an infrastructure most of American mining does not have.
Now look at where autonomy is actually scaling in the United States.
Aggregates. Quarries. Bridgeport is a limestone quarry. Bull Run is a crushed stone plant. Caterpillar has said openly that Bull Run does not use as many MineStar features as its large mining customers, that the solution was simplified to fit the site’s economics, and that training was trimmed to match. That is not a criticism. That is exactly how you make the technology viable for a quarry. It is also an honest admission that the version of autonomy arriving in aggregates is a leaner version of the system that built the record.
And a quarry is not the Pilbara.
I have spent my working life in these operations. The pit is tighter. The haul is shorter. The traffic is mixed all day, every day: haul trucks, loaders, water trucks, service vehicles, contractors, vendors, and customer trucks rolling through the scale house. There is no ops center in another city. There is a supervisor with a radio, a maintenance crew stretched across the plant, and a workforce where one person wears four hats. The berm gets rebuilt by whoever is on the dozer that week. The survey updates when it updates.
So hold both facts at once. The environment where American autonomy is scaling fastest is the environment with the least of the infrastructure that produced the clean record. Tighter pits, more mixed traffic, leaner staffing, simpler system builds.
That is not an argument against quarry autonomy. Bridgeport and Bull Run prove it can run clean, and keeping a quarry worker out of a haul truck matters exactly as much as keeping an iron ore operator out of one. It is an argument about where the margin is thin. Every serious autonomous incident on record happened where the machine world and the human world touched. Quarries are almost nothing but touch points.
The industry press is celebrating the milestone tonnage. Fine. Somebody also needs to say that the mixed traffic problem the Pilbara engineered away with segregation is the daily operating condition of an American quarry, and the safety program that manages it is being written site by site, right now, mostly without a template.
V. The liability collision
When something finally goes wrong, the money and the blame do not flow down one clean channel. They collide.
Start with the operator. Primary responsibility under MSHA and premises liability. For employees, workers’ comp is the exclusive remedy in most states. No fault, capped, and it generally bars the employee from suing the employer. That part is settled.
Except think about who is actually standing in a quarry. Contractors. Vendors. Haul truck drivers from customer companies. None of them are inside the operator’s workers’ comp bargain. They can sue. And when they sue, they do not stop at the mine.
Now the truck maker is in it. The software developer is in it. The integrator and the retrofit vendor are in it. The theory is product liability: design defect, failure to warn about the limits of the system, marketing that promised more capability than the machine safely delivered in the rain on a shifted berm. Automotive and warehouse robotics have already built the early case law a mining plaintiff’s attorney will borrow.
And here is a twist worth naming. The industry’s own safety statistics will be evidence.
Caterpillar reports more than 8.6 billion tonnes hauled autonomously with no reported injuries, across more than five hundred fifty trucks on three continents, with more autonomous distance traveled than any automobile manufacturer. Today those numbers are marketing. The day after an incident, they are exhibits. A defense attorney will hold them up to show the system is proven and the failure was a true edge case. A plaintiff’s attorney will hold up the same numbers to argue the opposite: with that much operating experience, this failure mode was knowable, foreseeable, and preventable, so why was it not prevented. Eight billion tonnes of experience is a shield right up until it becomes a measuring stick.
The deepest fight will be over the data log.
Discovery becomes a war over training data, validation testing, and integration decisions that everybody made together and nobody documented cleanly. Whose model made the call. What the truck saw. Whether that day’s behavior was a known limitation buried in a release note or a genuine surprise. The contracts written today mostly do not answer these questions, because they were written to buy iron, and what got delivered was iron with a brain that belongs to somebody else.
A quarry buying a retrofit autonomy package is signing up for this fight with less legal firepower than a Tier 1 miner. Which makes the contract, the indemnity language, and the data access terms the most important safety documents on the property that nobody thinks of as safety documents.
Aviation and nuclear already solved a version of this. Shared liability structures. Mandatory data standards agreed before the incident, not after. Black box requirements everyone can read. In a system where causation is genuinely shared, pure fault chasing produces bad safety outcomes and worse litigation. Mining autonomy will need its version of the same. You want those rules written on a calm day. Not reverse engineered by a jury.
VI. What the record from overseas actually teaches
Texas has a year on the board. The Pilbara has more than a decade, and the longer record is the gift, because it is long enough to include the bad days. Ours is not, yet.
The upside is documented. BHP has reported a reduction of more than eighty percent in significant events year on year tied to automation and related technologies. Fortescue’s autonomous trucks logged more than twenty four million kilometers safely since 2012. Rio Tinto runs roughly eighty percent of its Pilbara haul fleet autonomously. Fewer humans exposed to haulage, which is where a brutal share of mining fatalities have always come from.
Now the bad days, because that is where the lessons live.
In 2019, two autonomous trucks collided at BHP’s Jimblebar mine. The cause was weather. Heavy rain deteriorated the road surface and one truck slid into the other. Nobody was hurt, in part because no workers were in the immediate area. Rain, a degraded surface, a machine outside its envelope. In a quarry, that same event happens fifty feet from a mechanic.
That same year, two autonomous trucks collided at Fortescue’s Christmas Creek mine after a dropout in the wireless coverage connecting trucks to the control center. Connectivity is not a convenience in an autonomous pit. It is a safety system, and when it degrades, the machine’s judgment degrades with it. Now ask how many American quarries have hardened, redundant network coverage across the whole pit. I already wrote a piece about the radio on my hip. The connectivity gap is not theoretical to anyone who has worked a bench.
And the one that should be pinned to every wall: a Western Australian significant incident report on a collision between an autonomous haul truck and a manned water cart. The control system directed the truck through an intersection that existed in the software but was never physically marked on the ground. The water cart driver did not know the truck’s assigned path. By the time each was aware of the other, the truck could not stop. Investigators pointed to inadequate change management between the digital map and the physical pit.
Mixed traffic. A control system that knew something the human did not. A gap between the map and the ground. Shared causation across a machine, a control room, and a person who was never given the information to protect himself. That is not an Australian problem. That is a Tuesday at any operation where the digital pit and the real pit are maintained by different people on different schedules.
One more thing the manned era never had to think about. The daily walkaround was not just compliance. It was a sensing network. Thousands of informal inspections a year, ears catching a bearing starting to sing, eyes catching the hose that was fine yesterday. Take the operator out of the cab and you do not just remove a driver. You remove the most experienced diagnostic instrument the machine had. Telemetry and condition monitoring replace some of it. The judgment layer, the twenty year ear that knows what wrong sounds like, walks out of the pit with the workforce that used to sit in the seat. I have written about who maintains the brain inside the iron. This is why the question is not optional.
The workers see all of this clearly, and their view belongs in the room, not in the appendix. Fewer operator seats, more technician and controller roles, different skills, real upskilling or real displacement. The MSHA rule already requires miner input into the safety program for a reason. The remote operator’s workload, their trust in the system, and their authority to stop it are safety variables, not HR footnotes.
VII. What to do now
Nobody has to wait for a rulemaking, and given the current regulatory direction, nobody should.
For operators, especially in aggregates. Treat autonomy as its own risk assessment, not a paragraph in the existing mobile equipment program. Map every point where the machine world touches the human world, because in a quarry that map is the whole property. Write vendor contracts with teeth: clear indemnity, defined data sharing, and guaranteed access to the logs before the day you need them. Engage your MSHA district on a calm day and agree on what a compliant digital examination looks like at your site. Do not let the first conversation about your autonomous preshift happen during an accident investigation.
For equipment makers, software developers, and retrofit vendors. Functional safety and explainability are not features to add later. Validate against the edge cases that already happened, weather, connectivity loss, mixed traffic, unmarked changes to the pit, because those are not hypotheticals, they are the historical record. Warn honestly about system limits. The failure to warn claim is the easiest to prevent and the hardest to defend. And when you simplify a system to fit quarry economics, document what was simplified and what the site takes on in exchange. That honesty protects everyone, including you.
Jointly. Use the standards work that already exists instead of starting blank. The Global Mining Guidelines Group published version two of its Implementation of Autonomous Systems in Mining guideline in 2024, built by miners, equipment makers, and technology vendors together, covering planning, operational readiness, deployment, and change management. GMG has also published work on applying functional safety to autonomous mining systems. None of it is law. All of it is the template a quarry deploying autonomy this year should be holding. Pilot autonomy sandboxes with regulators. Standardize digital examination protocols so a driverless preshift means the same thing at every operation.
And push the early ideas before they are needed. Third party audits of safety critical AI, the way we audit financials. Digital twins that stress test a deployment before it moves real rock in front of real people. Hybrid oversight that keeps a human meaningfully accountable for the decisions that can kill someone. A Mine Autonomy Safety Board with operators, equipment makers, unions, and MSHA at one table before the first big case forces the conversation from a courtroom.
VIII. Shaping the brain inside the iron
I helped build one of these machines. I was on the team at Caterpillar that built the MD6 drill. That drill runs patterns today without a person in the cab. I know what goes into making iron think, because I was in the building when we did it.
And I have spent years in the quarries where this technology is arriving next, writing the preshift records the rules still assume a human fills out.
Sitting at that intersection is the whole point. We built machines that operate without the person. We kept a rulebook written around the person. And the agency that holds the rulebook has told us, in effect, that its rules attach where the people are, while the whole promise of the technology is to take the people away.
Autonomy is worth having. The safety case is real. The clean year in Texas is real. Keeping a driver out of the seat of the most dangerous machine in the pit is a genuinely good thing, in a quarry every bit as much as an iron ore mine.
But nobody is writing the autonomy rulebook. Not MSHA, whose agenda points the other direction. Not Congress. The rulebook is currently scheduled to be written by the first serious incident, in citations stretched over 1985 language and discovery fights over a proprietary data log.
Mining has absorbed explosives, diesel, and hydraulics, and wrote the safety culture to match each one. It can do it again, on purpose, in daylight, with the workers in the room.
The brain is already in the iron. The iron is already in the quarry down the road.
The only open question is whether we decide who answers for it before the day somebody has to.
Disruptive Depths, ai in heavy industry. Operations leader in mining and aggregates, translating AI into practical reality for heavy industry.
Glossary
Autonomous haulage system (AHS): A fleet of haul trucks operating without a driver in the cab, navigating defined routes and coordinated through a central control system.
Retrofit autonomy: An aftermarket system added to existing trucks rather than built in at the factory. The dominant model in quarry deployments.
Mixed traffic: A work area where autonomous and manned equipment operate together. Nearly every serious autonomous mining incident on record happened here. In quarries it is the default condition, not the exception.
Edge case: A condition the system was not well trained or designed for. Weather, unmarked terrain changes, and connectivity loss are the recurring real world examples.
Data log: The machine’s internal record of what it sensed and decided. After an incident it becomes the central, and usually proprietary, piece of evidence.
Performance based standard: A rule that sets an outcome and lets the operator choose the method, versus a prescriptive standard that dictates the method. MSHA’s mobile equipment rule is performance based, which is why it stretches over autonomy without answering autonomy’s specific questions.
Sources and notes
Regulatory framework, verified against primary sources:
MSHA, Safety Program for Surface Mobile Equipment, final rule, 30 CFR 56.23000 / 57.23000 / 77.2100. Published December 20, 2023, effective January 19, 2024, enforcement from July 17, 2024. Federal Register, msha.gov.
30 CFR 56.14100, Safety defects; examination, correction and records. eCFR, Cornell LII.
30 CFR 56.14207, Parking procedures for unattended equipment. eCFR, Cornell LII.
30 CFR 56.18002, Examination of working places.
MSHA enforcement posture on autonomous vehicles, including the close proximity view and the absence of autonomy specific regulation, as reported in mining trade legal commentary (Pit and Quarry).
MSHA’s July 2025 publication of eighteen proposed rules under the deregulatory executive order framework. Federal Register, Jackson Lewis analysis.
U.S. deployments:
Heidelberg Materials, Lake Bridgeport Quarry, Texas. Pronto AI autonomous haulage on a mixed Caterpillar and Komatsu fleet. More than two million tons moved driverless in under eight months, described as North America’s first fully autonomous mixed fleet quarry operation, with commitment to more than 100 self driving trucks globally by 2028. Heidelberg Materials, Pit and Quarry, Construction Equipment.
Luck Stone Bull Run Plant, Virginia. Autonomous Cat 777s, go live November 2024, more than two million tons in the first year on a single shift with no reported safety injuries. Caterpillar’s account notes the MineStar solution was simplified to fit the site’s requirements and economics. Caterpillar, Equipment World.
Caterpillar Command for hauling fleet record: more than 8.6 billion tonnes autonomously hauled with no reported injuries, more than 550 trucks across three continents. Caterpillar.
International record and incidents:
BHP Jimblebar autonomous truck collision, March 2019, heavy rain and deteriorated road surface. Australian Mining, Mining Technology.
Fortescue Christmas Creek autonomous truck collision, 2019, linked to a wireless connectivity dropout. MINING.COM.
Western Australia Significant Incident Report 226, collision between an autonomous haul truck and a manned water cart, unmarked intersection and inadequate control system change management. WorkSafe WA.
Rio Tinto Pilbara fleet scale, BHP significant event reduction figure, Fortescue kilometers figure. Company statements and trade reporting.
Standards:
Global Mining Guidelines Group, Implementation of Autonomous Systems in Mining Guideline, Version 2, 2024, and related GMG functional safety and system safety publications for autonomous mining. gmggroup.org.
Open sourcing note: stakeholder and worker perspectives are written as analysis. Named quotes from MSHA, operators, equipment makers, union representatives, and liability counsel should come from direct interviews before publication.