McKinsey’s mining outfit put out a report on AI making mines safer, and most outlets will run it as a feel-good tech story. I’ll do you the favor of cutting through it.
The core facts are real enough. Mining still kills people at rates that shame construction, manufacturing, and transport. McKinsey’s authors — Fagundes, Agarwal, and Yanto — say safety transformation has been uneven, and they’re right. But here’s what this really means: the barriers they list aren’t technology problems. They’re management problems wearing a technology costume.
Take the first one. Fragmented organizations, unclear accountability, corporate safety frameworks that don’t survive contact with the pit face. Translation: executives write policies in boardrooms and site managers quietly ignore them because production quotas pay the bonuses. McKinsey notes operators get “anchored in a compliance-driven, lagging-indicator mindset” — obsessing over audits and documentation instead of actually finding hazards. Anyone who’s sat through a corporate governance rollout knows exactly what this looks like. You can’t fix that with a neural network.
Second barrier: data silos and legacy infrastructure. Mining companies have spent a decade bolting sensors onto everything — slope stability monitors, equipment health telemetry — and then not connecting the data to anything useful. McKinsey’s example is telling: slope stability data that never makes it into company-wide dashboards. That’s not an AI gap. That’s a company that bought hardware for the investor deck and never budgeted the integration. I’ve watched this movie in finance too — firms buying trading infrastructure for optics and running the real book on spreadsheets.
Third: the critical minerals rush. Here’s a point McKinsey makes that deserves more airtime than it’ll get. The global scramble for copper, lithium, and rare earths has pushed rapid workforce expansion and heavy contractor reliance, especially in emerging markets. Inconsistent training, poor fatigue management, thin frontline supervision. When governments subsidize speed, safety is the first line item to get quietly cut. If you’re allocating capital to mining equities, contractor safety management in Tier-2 jurisdictions is a genuine tail risk — one fatalities scandal or tailings failure and your equity takes a 30 percent haircut overnight. This isn’t theoretical. Ask the investors who held Brumadinho-exposed names.
Fourth: change management. Pilots in autonomous haulage and fatigue monitoring show benefits, but scaling requires leadership alignment, workforce adoption, upskilling. Every technology consultant in history has written this paragraph, and it’s true every time: the pilot works, the rollout stalls, and the CFO quietly defunds it in year two.
Now the technology itself. Since around 2010, autonomous trucks and remote monitoring have spread through the industry, and this part I’ll credit — automation removing humans from harm’s way is the one safety intervention with a decades-long, unambiguous track record. Fewer people underground means fewer people underground when things go wrong. Simple arithmetic.
McKinsey lays out four levers. Drones and IoT sensors controlling hazards — mapping shafts, tailings dams, blast zones, monitoring air quality and ground stability in real time. Centralized control rooms aggregating live data into safety dashboards. Wearables and dash cameras for fatigue and PPE monitoring — with consent and privacy guardrails, they note, which is doing a lot of work in that sentence. And digital twins for training — simulated stress scenarios before you touch the real mine. All reasonable. None of it is new. The sensors and the autonomy have existed for years; the report’s actual news is the generative and agentic AI layer on top.
And here’s where I get skeptical, because this is where the consultant economics kick in.
Gen AI copilots summarizing safety data and drafting incident reports in seconds — fine, that’s paperwork automation dressed up as safety innovation. LLMs mining thousands of near-miss reports to find recurring precursors like overridden alarms or contractor interface breakdowns — this is actually the most credible use case in the whole report. Near-miss data is text-heavy, siloed, and chronically under-analyzed because nobody has time to read it. If a language model surfaces patterns humans missed, that’s genuine value. Just don’t expect miracles; the model is only as good as the incident reports, and those reports are only as honest as the site culture allows. Garbage culture in, garbage insights out.
The “agentic AI” piece — autonomous systems adjusting ventilation and rerouting haul trucks around unstable ground — this is where I’d slow down. McKinsey dutifully notes human oversight is always required, and thank God for that, because the day you let an autonomous agent make safety-critical decisions at a mine site is the day your general counsel earns his salary. Regulators will not accept “the model decided” as an incident explanation, and neither will juries. Autonomy within tightly bounded, pre-defined parameters? Already happening, sensible. Free-ranging agentic systems underground? That’s a PowerPoint slide, not a deployment plan.
The case studies are the usual anonymized consultant fare. A coal operator with a domain-trained model integrating video, sensors, and text to flag restricted-zone intrusions. A transshipment operator using video analytics on existing CCTV — crew slips, engine room smoke — routed to human analysts who verify before escalating, with an AI agent drafting structured incident reports afterward. A global miner building AI-backed contractor management frameworks. These are credible deployments. Notice the pattern, though: in every case, a human verifies, a human decides, a human evaluates. The AI is a very fast junior analyst, not a decision-maker. That’s the honest framing, and it’s the frame the press release version tends to blur.
McKinsey closes with the line that intelligent safety is becoming “a foundation of competitive advantage” — safer mines are more productive mines, less unplanned downtime, better asset utilization. This part I actually agree with, though not for the reason consultants like. Safety and operational discipline are the same muscle. A company that can’t keep its sensors integrated, its contractors trained, and its near-miss reporting honest isn’t going to be saved by an AI platform. The AI amplifies whatever operating culture already exists. In a disciplined organization, it compounds. In a sloppy one, it automates the sloppiness.
So what’s the practical read if you’ve got capital in this sector? Three things. First, treat safety technology adoption as a proxy for management quality — companies that integrated their sensor data years ago tend to be the ones that execute on everything else too. Second, watch contractor-heavy, emerging-market critical minerals operations; that’s where the capability gap McKinsey describes creates real downside. Third, discount any miner’s AI narrative until you see scaled deployment with auditable safety metrics, not pilots. This industry has seen every hype cycle roll through — drone mania, blockchain supply chains, metaverse training sims. The tech eventually finds its proper, modest place. The circus leaves town, the sensors keep humming, and the disciplined operators quietly compound. That part never changes.
Source: https://im-mining.com/2026/11/18/mckinsey-on-harnessing-ai-to-make-mining-safer/

