Two workers survived nine days in a Nepal tunnel: what the rescue teaches about emergency engineering

Two men came out alive after nine days

On 4 September 2026, teams cutting through mud, rock and buried structures found life where objective signs had become scarce. Sanjay Sah, described by Reuters as a mechanical foreman, and Kabir Maharjan, a mechanical supervisor with the Nepal Electricity Authority, were rescued from an underground structure linked to Upper Trishuli 3A.

They had remained underground for nine days after the 26 August floods. The engineering point is not to turn survival into spectacle. It is to understand what allowed the search to continue after the normal exit disappeared, the site's geometry was buried and the rescue itself generated new hazards.

An emergency plan does not end with an evacuation order. When evacuation becomes impossible, alternative access, technical information, communication, personnel location and rescuer safety determine what can still be done.

The disaster is context; emergency engineering is the subject

The floods damaged homes, roads, bridges, hydropower facilities and critical infrastructure across Rasuwa, Nuwakot, Dhading and other districts. Andrade Safe's earlier Nepal analysis covers early warning, evacuation, redundancy, first aid and responder protection. This article begins where normal evacuation failed or was no longer possible.

Nepal's 10 September bulletin reported 13,656 people rescued, 5,130 missing and 1,377 bodies recovered. The figures remain subject to verification. Search and rescue continued at projects including Upper Trishuli 1, 3A and 3B, Chilime, Rasuwagadhi, Mailung Khola and Langtang.

The 9 September bulletin had specified three live tunnel rescues: the two Nepali nationals at Upper Trishuli 3A and one Chinese national at Upper Trishuli 1. The 10 September edition did not repeat that breakdown, so three remains the latest official tunnel-specific total rather than a newly reported count.

Read the broader disaster and emergency-preparedness analysis →

Official fact, journalistic reconstruction and engineering lesson

The government confirmed the two rescues and the continuing operations. Most operational detail comes from a Reuters reconstruction based on interviews, maps, documents and internal rescue conversations. It is high-quality reporting, not an official investigation report.

Drilling, controlled blasting, pressurised water, an escape ladder and communication difficulties may therefore be described as reported elements. They cannot be converted into an official cause, a legal breach or a universal rescue recipe.

A technical question is not an accusation. It is the path from an event to evidence.

An air-filled space, without inventing the atmosphere

Reuters reported that Sah and Maharjan reached a cable-access tunnel with an air-filled chamber about 160 metres from the exit. Maharjan reportedly drank muddy water. These human details do not provide a complete physiological explanation.

No public measurements establish oxygen, contaminants, temperature or water quality at their location. It would be wrong to call the atmosphere safe, to say supplied air guaranteed survival or to identify one factor as the reason they lived. Underground atmospheres require measurement, not visual assumptions.

When the map no longer resembles the site

Reuters described four access routes blocked by mud, rock, timber and debris, with parts of the facility under 15 to 30 metres of landslide material. The physical reference that existed before the flood had become a new, saturated and unstable landscape.

As-built drawings, levels, shafts, galleries and cable routes can then become rescue resources. But information that cannot be retrieved during an emergency is worth less than it appears in the archive. Current drawings must be available offline and outside the likely impact zone.

A drawing still is not the transformed site. It guides the search, while field evidence must confirm what survives and where it is.

Alternative access and the cable gallery

Local reporting and Reuters indicate that a cable-access tunnel played a central role. A technical route assumed an unexpected emergency function after primary approaches were buried.

That does not make every cable tunnel an escape route. The lesson is to know the asset's full geometry and the alternatives that might remain after water, mud, loss of power or deformation. Redundancy is not duplication for paperwork; it is another option when the primary route disappears.

Stabilise first, then move

The ground was reportedly too saturated for heavy equipment to move in immediately. Crews first established a stable platform. An excavator only accelerates a rescue if the ground can support the excavator.

Urgency does not remove bearing capacity, line-of-fire, rollover or secondary-collapse hazards. What looks like delay may be the control that preserves machines, access and people for the next operational period.

Drilling to turn uncertainty into information

Reuters described drilling, controlled blasting and a roughly five-inch borehole used to confirm the buried tunnel's alignment. Its value was not merely opening material; it reduced uncertainty about where to focus limited effort.

Exploratory drilling can inform position, material and access, but method, diameter, energy, ventilation, stability and the risk of striking people or services require specialist design. This event is not an operating instruction for drilling or blasting.

Water, rain and collapse: the hazard kept changing

Reuters reported that pressurised water through a natural channel shifted part of the final obstruction. That outcome cannot be generalised as a recommended tunnel-rescue technique.

Later rain triggered a partial collapse and briefly trapped rescuers, according to the report. The passage had to be reinforced. A control can change drainage, ground behaviour and stability; conditions must be reassessed continuously.

Rescue is a second high-risk operation

Rescuers may face unstable ground, water, mud, altered atmosphere, residual energy, heavy equipment, restricted passages, fatigue and new collapse. Saving someone does not justify creating another victim.

A mature operation defines entry and withdrawal criteria, communication, tracking, support, stop authority and a plan to rescue the rescuer. These are general principles, not claims that specific controls were absent at Trishuli 3A.

Stopping the machines to listen

Reuters reported that engines were repeatedly stopped so teams could listen for signs. Excavation productivity and survivor detection required different operating states. The same noise that moved debris could conceal the most important information.

A search may require the operation to pause, become quiet, measure and observe before resuming. Near dawn on the tenth day, uncertain sounds became a direction for the final approach.

Drawings, messages and asset memory

A separate Reuters report described engineers exchanging drawings, photographs and hypotheses through a WhatsApp group. The platform is not the central lesson. What mattered was access to people who understood the installation and could turn technical information into usable decisions.

Critical knowledge cannot depend on one employee, one inaccessible room or one network connection. Emergency information packs should include as-builts, access routes, energy isolation, hazards and specialist contacts in controlled offline copies.

The escape ladder remains a question, not a verdict

Reuters reported that rescuers later identified an escape ladder and quoted an NEA engineer saying workers did not know where to go. No public official report had established whether the route was accessible, marked, trained or usable during the flood at the time of this review.

It is not responsible to conclude that the route failed, training was absent or the ladder would have saved everyone. The legitimate question is how an organisation makes an exit operationally real: known, recognisable and reachable under stress and physical transformation.

Personnel accountability and underground communication

Without a reliable account of who entered, who left and where each team was working, rescue begins with another uncertainty. Employees, contractors, visitors and shift records must converge into an operational count.

Underground communication also needs redundancy. Conventional radios may lose coverage, and power or networks may fail. Plans need environment-compatible primary and backup methods, contact intervals and a clear response when silence becomes an alarm.

What OSHA and ILO frameworks help us see

OSHA 29 CFR 1926.800 addresses underground construction and recognises access control, check-in/check-out, communications, ventilation, lighting, water-inrush protection and rescue availability. It is not Nepalese law and does not prove a breach in this case.

ILO Convention 167 and the revised construction code provide a broader framework for excavations, shafts and tunnels. They support preventive questions about location, communication, atmosphere, egress and rescuer safety without turning international guidance into a finding against the project.

Brazilian comparison: NR-1 and caution with NR-33

Brazil's NR-1 requires emergency response procedures coherent with applicable scenarios, resources and responsibilities. It is useful for Brazilian organisations reviewing loss of access, communication failure, evacuation and responder protection.

NR-33 does not automatically classify every tunnel as a confined space. Classification depends on the real characteristics and regulatory criteria. Neither Brazilian standard applies legally in Nepal; both are used only as technical comparisons.

Educational underground-rescue readiness checklist

This list does not replace risk assessment, egress or ventilation design, an emergency plan, rescue procedures or applicable law.

• Current inventory of primary and alternative access and egress.
• As-built drawings available quickly and offline.
• Reliable personnel accountability and location system.
• Environment-appropriate primary and backup communication.
• Evacuation triggers and authority to order and confirm withdrawal.
• Energy-zone identification and isolation strategy.
• Water, mud, drainage and ground-stability assessment.
• Atmospheric assessment, ventilation and emergency lighting where applicable.
• Specialists who know the asset's geometry.
• Equipment compatible with access and ground capacity.
• Rescuer entry, withdrawal and rescue-of-rescuer criteria.
• Exercises that test loss of the primary route and post-change review.

Conclusion: prepare for the moment the ideal route disappears

Nine days later, two workers emerged alive from a facility overtaken by mud, water and debris. The response did not come from one technology. It combined information, ground reading, access, drilling, listening, adaptation and teams willing to change strategy as the risk changed.

The latest reliable reporting described Sah as stable with non-critical injuries. Reuters reported on 8 September, citing Maharjan's wife, that he underwent surgery for a deep infected leg wound and was initially disoriented. No later reliable clinical update was found by 10 September.

An APF team said on 4 September that it found no one else in the sections it reached at Trishuli 3A; later government bulletins still included the project in wider search and rescue operations. These may refer to different areas or scopes.

A mature emergency plan prepares not only the ideal exit. It also prepares the organisation for the moment that route no longer exists, without turning the people who enter to save others into new victims.