MSHA's historical tables list 8 coal mining fatalities and 25 metal and nonmetal fatalities for 2025, against 10 and 18 for 2024. In its 2025 safety alert, MSHA said ten miner fatalities occurred between 3 January and 5 March 2025, "more than triple the number for the same period in 2024." BLS puts the 2024 fatal injury rate for underground mining machine operators at 35.6 per 100,000 full time equivalent workers, against 3.3 for all workers.
Those numbers are why mines look hard at remote operation and automated machine safety. Fewer people near a moving machine or an unstable face is a reasonable goal. It also moves risk into software, and software has to be treated like safety equipment. This post is about that software side. It does not claim any product reduces injuries.
What's affected
Three kinds of mining software carry safety weight:
- Remote operation links, where an operator drives a machine from a control room or surface station.
- Machine safety functions, such as proximity detection that stops a machine.
- Fleet and monitoring systems that watch machines and the environment, and increasingly use AI to flag problems.
MSHA's alert shows where the hazards sit. Of the ten fatalities, powered haulage accounted for four, and four "involved improper maintenance or unsafe operation of equipment." One involved "not locking and tagging out equipment." The alert does not say remote operation would have prevented any of them, and this post does not either.
What the software must get right
A rule in the US coal regulations shows what a safety function in software looks like. Under 30 CFR 75.1732, continuous mining machines (other than full face machines) must carry proximity detection systems. The rule requires the system to stop a machine that is tramming or repositioning before it contacts a miner, to warn the miner first, and to prevent movement if a machine mounted component is faulty. Each shift, a designated person checks the components, and the checks, defects and corrective actions are recorded in "a secure book or electronically in a secure computer system not susceptible to alteration," and kept for at least one year.
Pull four software requirements out of that rule:
- Fail safe. A faulty component stops movement. It does not allow it.
- Safe stop. The machine stops on its own logic, not on a command from a remote system that may have lost its link.
- Tamper resistant records. Logs that cannot be quietly edited.
- Known good state. Someone checks at the start of each shift that the system works.
Remote operation adds three more:
- Secure, authenticated updates. A firmware change on a machine controller is a safety change. It needs signing and a record of who approved it.
- Offline operation. Underground, the network is intermittent. The system must behave safely when the link drops, and must not depend on a cloud service to keep a machine stopped.
- Least privilege for remote sessions. Who can issue a motion command, from where, and with what approval.
What to do now
- Inventory the software on each machine controller and remote link, including versions and who maintains them.
- Match that inventory to known vulnerabilities, giving priority to anything on the CISA Known Exploited Vulnerabilities list.
- Require signed updates and keep a log of every update applied to a safety related controller.
- Test the link loss behaviour on purpose, with the machine stopped and supervised.
- Run scanning and any AI assistance inside the site network when connectivity is limited. Safeguard supports on premises and fully air-gapped deployment, including offline models.
Quick check
What does 30 CFR 75.1732 require when a proximity detection component is faulty?
The system must prevent movement of the machine. Movement is allowed only to relocate it from an unsafe location for repair, with a distinguishable warning.
