Will AI Replace Subway Operators? The Underground Automation Debate

The question is not academic. In Paris, Line 1 of the Metro has carried passengers without operators since 2012. In Copenhagen the entire metro network is driverless. Singapore, Dubai, Vancouver, and parts of Hong Kong all run substantial portions of their rail networks without human operators in the cab. Yet New York, London, Tokyo, and dozens of other major systems still have human operators on every train. So who is right?

The data we use to think about this question for individual workers gives a clear answer for subway operators: 55% automation risk with 42% AI exposure. That risk score is among the higher ones we see for skilled transit roles, and it correctly captures the fact that the technology to automate subway operation has existed for decades and has been deployed successfully in many cities. But it also undershoots reality in some cities and overshoots it in others, because the answer depends heavily on infrastructure age, labor agreements, and political dynamics.

This article unpacks what is actually happening to subway operator jobs in 2025, why the picture varies so dramatically by city, and what an operator should be thinking about over the next ten years. The data here draws from O*NET task analysis, the International Association of Public Transport (UITP) automated metro statistics, and labor market reports specific to the transit sector.

What 55% Risk Means in a Profession Like This

Subway operation is unusual among the occupations we analyze. The technical work — driving the train, opening and closing doors, communicating with control — is genuinely automatable, and has been for forty years. London's Victoria Line has had Automatic Train Operation since 1968, with operators serving as supervisors rather than drivers in the conventional sense.

What stops automation is not technology. It is some combination of:

Legacy infrastructure costs. Retrofitting an existing subway line for driverless operation typically costs $200-400 million per route-kilometer, including signaling upgrades, platform screen doors, and integrated control systems. For systems built before 1980, the rolling stock, signaling, and station infrastructure are often incompatible with driverless operation without massive replacement. The capital expenditure required is more than most cities can justify when human-operated service already works. [Fact]

Labor agreements. Many transit unions in North America and Europe have negotiated contractual provisions that effectively require human operators. The Metropolitan Transportation Authority in New York City and Transport for London both face strong union opposition to driverless conversion. These agreements can be renegotiated, but doing so takes years and often political capital that elected officials are unwilling to spend.

Public acceptance. Driverless metros are normal in cities that built them that way (Paris Line 14, Copenhagen, Dubai). They are politically charged in cities that already had operators. Conversion proposals frequently face public opposition that delays or kills them, even when transit agencies want to proceed.

Emergency response complexity. In normal operation, automation handles the work easily. In emergencies — fires, security incidents, medical events, signaling failures — the value of having a trained human on the train rises dramatically. Many cities decide the insurance value of having operators on trains justifies the staffing cost, even when normal operation does not require them.

So the 55% risk score captures the long-run trajectory accurately while glossing over the fact that the trajectory plays out over decades, not years, and that some operators will retire in current positions while others are facing imminent automation.

The Cities Where Operators Are Already Gone

The list of fully or substantially driverless metros has grown steadily. As of 2025 the UITP reports more than 75 driverless metro lines operating in over 40 cities worldwide. The trend is accelerating: new lines are increasingly designed without operators from day one, and selected conversions of existing lines are completing every year. [Fact]

Cities where automation has either taken over or is well advanced include Paris (Lines 1, 4, 14), Copenhagen (full network), Dubai (full network), Singapore (Northeast, Circle, Downtown, Thomson-East Coast Lines), Hong Kong (Disneyland Line, South Island Line), Vancouver (SkyTrain network), Sao Paulo (Line 4), Santiago (Line 6), Doha (Doha Metro), and Riyadh (recently opened Lines 1-6).

What is striking about this list is that almost every entry represents either a new-build system or a planned conversion that took many years to execute. Cities with legacy systems that were not designed for driverless operation tend to make incremental modifications rather than full conversion. New York's Canarsie Line (L train) installed Communications-Based Train Control but kept human operators. The Washington Metropolitan Area Transit Authority has discussed automation for years without committing to full conversion.

The Cities Where Operators Are Likely Staying

Some major systems will probably keep human operators for the foreseeable future, for reasons that are part technical, part political, and part economic.

New York City Transit. The largest subway system in North America carries 5.5 million passengers per weekday across 472 stations and 27 lines. Retrofitting the entire system for driverless operation would cost tens of billions of dollars and decades of construction. Union opposition is strong. Public skepticism is high after several years of subway crime concerns. Operators here are probably safe through at least 2040.

London Underground. The system has been gradually adopting Automatic Train Operation on individual lines (Jubilee, Northern, Victoria, Central) but keeps operators in cabs in supervisory roles. Conversion to fully driverless has been proposed repeatedly and rejected each time by Transport for London leadership. The current direction is more automation with human supervision, not human removal.

Tokyo Metro and Tokyo Toei. Despite Japan being a technology leader generally, Tokyo's subway operators are heavily unionized and operations are conservative. The systems are running near capacity already, which limits appetite for disruption. Driverless conversion is not on the near-term agenda.

Most North American systems. Boston, Chicago, Philadelphia, San Francisco, Toronto, Montreal, and others all face similar combinations of legacy infrastructure, labor agreements, and political dynamics that make near-term driverless conversion unlikely.

The Tasks AI Is Affecting Today

Even in cities where operators are staying, AI is changing parts of the job.

Predictive maintenance alerts. Modern subway operators now receive real-time information about train health — wheel temperature anomalies, door cycle wear, propulsion irregularities. AI systems generate the alerts and the operator's role shifts from monitoring instruments to responding to identified issues. This is a productivity gain for the operator and a safety gain for passengers.

Schedule and route optimization. Dispatching decisions about train movements during disruptions are increasingly AI-assisted. The operator receives clear instructions through a Transit Control Center that is itself relying on AI tools to model the cascading effects of decisions.

Passenger information and announcements. Routine station announcements, delay explanations, and route information are increasingly automated. The operator's voice-over-passenger announcements have been largely replaced by recorded or AI-generated audio in most modern systems.

Documentation and incident reporting. When events happen — medical emergencies, security incidents, mechanical issues — the operator now uses tablet-based reporting tools that pre-fill standard fields and suggest classifications. AI handles much of the documentation burden.

Training simulators. New operators learn on AI-enhanced simulators that present scenarios drawn from real incidents across the network. Training is more comprehensive and standardized than the classroom-and-shadow approach of the past.

The Tasks That Remain Human

The reasons operators still exist in non-driverless systems boil down to a specific set of tasks AI cannot perform well.

Emergency response. When a train must evacuate in tunnel due to fire, smoke, or security threat, the operator is the on-scene authority for hundreds of passengers. They coordinate with emergency services, manage evacuation routes, and make second-by-second judgment calls about passenger safety. There is no AI replacement for this work.

Mechanical troubleshooting. When a train fault prevents normal operation, the operator often performs initial diagnosis. They cycle systems, attempt manual operation, and communicate detailed observations to maintenance crews. This work is hands-on and judgment-dependent.

Passenger interaction. Medical events, intoxicated passengers, conflicts between riders, lost children — the operator is often the on-scene authority for handling these situations until other staff arrive. Calm professional judgment under stress is genuinely valuable.

Signal anomaly response. When trackside signals or block protection systems behave unexpectedly, the operator slows or stops the train and verifies the situation visually. This is exactly the work humans do better than automation in the current generation of train control systems.

Communication during disruption. When normal operations break down, the operator speaks with passengers, with control, and with emergency services. Translating between these audiences requires judgment about what to say, when, and how. AI is bad at this.

Manual operation. When automatic systems fail, qualified operators can drive trains manually under degraded service rules. This is the most important fallback in any automated system, and it requires a trained human who has not let their manual skills atrophy.

Career Path Implications for Subway Operators

For an operator reading this article, the relevant questions are: how secure is my current job, and what should I be doing about my career?

If you work in an existing legacy system without committed automation plans (New York, London, Boston, Chicago, Tokyo, and most North American systems), your job is probably secure through your retirement. Unions are strong, conversion costs are prohibitive, and political will for driverless service is weak. You may see your role evolve toward more supervision and less direct driving, but the position will exist.

If you work in a system that is currently being converted (selected lines in major North American and European cities), you should be paying attention to whether your employer is offering transition support. Many conversions retain operators in modified roles — train attendants, customer service ambassadors, control center staff. Negotiating for these transition paths is the relevant union and personal priority.

If you work in a system that is being designed with automation in mind (new builds and recent expansions), your career was always going to involve more transition. Operators in these systems often move into supervisory, training, or maintenance roles after a few years on the trains.

If you are considering subway operation as a career in 2025, the answer depends on where you live. In a legacy system with strong unions and no automation plans, it remains a stable middle-class career with strong benefits. In a new-build city or one with aggressive conversion plans, the career timeline is shorter and career planning more important.

The Honest Long-Term View

By 2040, perhaps 60-70% of subway operator hours globally will have been automated, with the remaining 30-40% concentrated in legacy systems in major cities. The transition will be uneven, with some cities going fully driverless and others maintaining human operators for safety, labor, and political reasons. The work itself, where it remains, will continue to evolve toward supervision and emergency response rather than direct train control.

For individual operators, the strategic message is to take advantage of the transition window. Develop skills that are valuable across the transit system — emergency response certifications, customer service training, technical maintenance literacy, supervisory experience. Position yourself to move into roles that will exist after your specific job is automated. Most importantly, engage with your union about transition agreements before automation arrives, not after. The operators who fare best in this transition will be those who are involved in shaping it.

For task-level automation breakdowns by sub-role, regional risk variations, and a detailed timeline of expected changes, see our Subway Operators occupation profile.

---

Analysis based on ONET task-level automation modeling, the Anthropic Economic Index (2025), International Association of Public Transport (UITP) statistics, transit agency public reports, and OECD AI Policy Observatory data. AI-assisted research and drafting; human review and editing by the AIChangingWork editorial team.*