The Autothrottle's Hidden Modes, the Asiana Two-Fourteen Accident at San Francisco, and the Mode Confusion Problem Reshaping How Aviation Trains Its Pilots

The 2013 Asiana 214 crash at San Francisco wasn't caused by bad pilots - it was caused by automation mode confusion, and the lesson reaches every cockpit with modern avionics.

Aviation Technology Analyst

On July 6, 2013, Asiana Airlines Flight 214 crashed short of Runway 28L at San Francisco International Airport in clear, unrestricted visibility, killing three passengers and injuring 182 others. The NTSB’s exhaustive investigation identified the initiating cause not as mechanical failure or crew incompetence, but as autothrottle mode confusion - a mismatch between what the crew believed their automation was doing and what it was actually doing. That failure mode has been documented in the aviation safety record for four decades, and it extends far beyond wide-body airline operations.

What happened on Asiana Flight 214?

The aircraft, a Boeing 777-200ER arriving from Seoul, was cleared for a visual approach to Runway 28L. During a configuration change on the descent, the autothrottle transitioned from speed mode into HOLD mode. The captain’s mental model, based on what he saw on his displays, was that the autothrottle was actively managing airspeed. It was not. The aircraft began decelerating. The energy bled off undetected.

A compounding factor: the ILS glide slope on Runway 28L had been NOTAMed out of service that day, forcing the crew onto a visual approach without precision vertical guidance. San Francisco’s bay environment creates a low-texture visual picture over water on short final, and the approach workload was elevated. The speed decay went undetected until the stick shaker activated at 125 feet above the ground. There was nothing left to do.

What is autothrottle HOLD mode and why is it easy to miss?

The autothrottle on a Boeing 777 operates in multiple distinct modes. Speed mode is the intuitive one: the system actively adjusts thrust to hold a target airspeed, adding power when the aircraft slows and pulling it back when speed climbs past target.

HOLD mode does none of that. The throttle levers freeze at their current position. The system stops adjusting for changing energy states entirely. Engines stay at whatever thrust level existed when the mode transition occurred, and airspeed becomes a free variable, dictated purely by aerodynamics and aircraft configuration.

HOLD mode is not a failure state - it is a designed operating mode with legitimate uses. The problem is what it looks like from the flight deck. The autothrottle remains engaged. The system appears operational. There is no prominent, unmistakable visual cue that speed management has stopped. The crew has to know what to scan and what they are looking at to catch the transition.

What is automation surprise, and how well-documented is it?

Automation surprise occurs when the automation transitions to an unexpected state - or stops doing something the pilot expects - without alerting the crew clearly enough to register the change. The mental model of what the system is doing diverges from reality, and the system doesn’t close that gap aggressively enough to demand attention.

NASA’s human factors division has been publishing on this problem since the mid-1980s, when the first full glass-cockpit airliners began generating incident reports. The finding has been consistent across four decades of data: the more capable the automation, the more modes it requires to deliver that capability. The more modes there are, the harder it becomes to track exactly which mode is active at any given moment.

Asiana 214 is not an outlier. In February 2009, Turkish Airlines Flight 1951 crashed on approach to Amsterdam Schiphol Airport, killing nine people. A Boeing 737-800 with a right-side radio altimeter stuck at negative eight feet caused the autothrottle to reduce power to idle - as designed, as if the aircraft were already in the landing flare. The crew did not catch the energy decay. The mechanism differs from Asiana 214, but the underlying failure is identical: automation operating in a state the crew wasn’t accurately tracking.

Does this problem exist across different aircraft architectures?

Both major manufacturers have built this problem into their systems, through no fault of the engineering. Boeing and Airbus hold genuinely different philosophies on how much authority computers should have versus the pilot, and reasonable people fall on both sides of that debate.

But mode complexity is an unavoidable consequence of automation capability in both architectures. The Airbus A320 family’s Flight Management Guidance and Envelope Computer (FMGEC) operates through several dozen distinct modes and submodes. Some share annunciator labels. Some behave similarly in normal conditions and very differently in degraded or edge-case scenarios. These are not design flaws - they are the cost of delivering the capability the systems provide. More automation means more modes. More modes means more surface area for a crew’s mental model to drift from reality.

How has the industry responded since 2013?

Boeing issued software updates to the 777 autothrottle system following Asiana 214. The changes improved alerting when the system transitions through modes that do not actively control airspeed, with the goal of making mode state changes more visible during a normal scan.

The FAA issued guidance to operators specifically targeting manual flight skills and energy management proficiency. The concern was stated explicitly: pilots who fly highly automated aircraft for years can allow foundational energy management skills to atrophy without recognizing it. Simulator evaluation data at the time indicated that manual proficiency was more degraded than crews realized.

The European Union Aviation Safety Agency (EASA) published a fleet-wide review that aligned with what the NTSB and NASA had already documented. Manual flying proficiency was inconsistent. Mode awareness was unreliable. Many experienced line pilots were unaware of either issue until a scenario exposed the gap.

ICAO published guidance for member states treating manual flight competency as a specific regulatory concern - not an abstract policy position - in direct response to data on what extended automation use does to pilot skill over time.

The industry has moved toward evidence-based training: simulator scenarios targeting the specific failure modes in the accident record, with automation surprise and mode confusion explicitly on the list. Major training organizations have been rebuilding curricula around this for the better part of a decade.

Why does this matter for general aviation pilots?

The autothrottle question is no longer exclusive to airline cockpits. Innovative Solutions and Support (IS&S) received FAA certification for the ThrustSense autothrottle system on the King Air series - the first time the King Air 350 had a certifiable autothrottle capability. It actively maintains airspeed on approach, manages torque in the climb, and protects turbines against overtorque exceedance. For single-pilot King Air operations in IMC, those are genuine safety improvements.

But ThrustSense has modes - auto mode, manual override, speed hold, torque hold - that behave differently from each other in ways that matter on approach. The same lesson from Asiana 214 applies directly.

The Garmin GFC 600, certified for the larger end of the general aviation turbine fleet and integrated with G1000 and G3000 suites, includes speed protection, envelope alerting, and visual approach modes. Its mode annunciator is live flight data, not background information.

The next generation of GA automation - predictive flight path management, automatic go-around initiation, envelope protection capable of overriding pilot inputs - will increase mode complexity before interface design catches up. The Garmin Autoland system already in service on certain aircraft demonstrates that fully automatic systems can land an airplane safely when the pilot is incapacitated. That is a real, meaningful safety capability. But every new function adds modes, and every new mode is another state where a mental model can diverge from what the airplane is actually doing.

How should pilots build mode awareness into their flying?

Know your automation’s modes before you fly. Not just the master switch location or the normal cruise configuration. Know what the system does when it transitions between modes. Know what the annunciator looks like in speed hold versus throttle hold versus a passthrough state. Know what the throttle servo does - and does not do - in each of those states. If you cannot describe what your autothrottle is doing in HOLD mode without looking it up, you have not fully learned the system.

Build the mode annunciator into your instrument scan. The PFD, engine instruments, and navigation display are standard scan items for most pilots. The autopilot mode strip is not background information - it is real-time data about what your automation is doing at this specific moment. Put it in your scan the same way you put the altimeter in your scan.

Fly without automation regularly. This is not preference or nostalgia - it is what the data supports. Pilots who hand fly regularly show better recovery performance in simulator evaluations involving automation failures and unexpected mode transitions. Manual flight keeps active the neural pathways that extended automation use allows to atrophy. The FAA and ICAO both recommend that instrument pilots hand fly a meaningful portion of approaches under actual or simulated IMC - not because automation is dangerous, but because if the automation surprises you, the reflexes need to be ready before the surprise arrives.


Key Takeaways

  • Asiana 214, July 6, 2013: The initiating cause was an undetected autothrottle transition to HOLD mode that removed active airspeed management during a visual approach in clear conditions.
  • HOLD mode looks operational from the flight deck: the system stays engaged, no prominent alert fires, and catching the transition requires knowing exactly what to scan for.
  • Automation surprise is a 40-year documented failure mode. Turkish Airlines Flight 1951 in February 2009 is a parallel example with a different mechanism and the same fundamental cause.
  • The FAA, EASA, and ICAO have all responded with guidance targeting manual proficiency and mode awareness as explicit regulatory concerns, not abstract policy positions.
  • The lesson applies to general aviation. Any pilot flying ThrustSense, the GFC 600, or similar systems faces the same cognitive demands - know your modes, include them in your scan, and hand fly regularly enough to stay ahead of the automation when it transitions into a state you weren’t tracking.

Radio Hangar. Aviation talk, built by pilots. Listen live | More articles