Asiana Two Fourteen, the HOLD Mode Nobody Monitored, and What the San Francisco Seawall Taught Aviation About Automation Complacency
Asiana Flight 214 crashed at SFO on July 6, 2013 not from mechanical failure, but because the crew didn't know their autothrottle had stopped controlling speed.
On July 6, 2013, a Boeing 777-200ER struck the seawall at the approach end of runway 28L at San Francisco International Airport on a clear afternoon with better-than-ten-mile visibility. Three passengers were killed. 187 were injured. The aircraft was not broken. The engines worked. The weather was perfect. The crew flew into the ground because they believed their autothrottle was maintaining approach speed - and it wasn’t.
What Was the Autothrottle Actually Doing?
The Boeing 777’s autothrottle operates in several modes depending on active selections on the mode control panel. One of those modes is HOLD mode - a state in which the autothrottle does not command thrust and does not maintain a target airspeed. It parks the throttle levers at their current position and does nothing further.
During the descent into SFO, the autothrottle transitioned into HOLD mode through a sequence of mode selections the crew did not fully register. The crew had been using FLCH (Flight Level Change) mode, which manages speed during climbs and descents. When the autopilot was disconnected for the hand-flown visual approach, the autothrottle’s behavior changed in a way the flying crew did not catch. The HOLD annunciation was displayed on the primary flight display - the indication was visible, not hidden.
Nobody saw it.
The throttles were at a low thrust setting. The target approach speed was approximately 137 knots. With the autothrottle parked and the engines at reduced power, speed began to decay.
How Did the Speed Decay Go Unnoticed?
The crew was flying the approach visually, monitoring the PAPI for glidepath. The speed decay was gradual - not dramatic enough to trigger an immediate alert. That gradual onset is part of what makes an undetected HOLD mode so dangerous: no sharp warning, just a slow erosion of energy state.
The flight data recorder documented what followed with precision. At approximately 500 feet AGL, airspeed was around 112 knots - 25 knots below the target approach speed. The crew was increasing pitch to control the descent rate, which at a decaying airspeed compounds the problem directly. At approximately 160 feet AGL, the stick shaker activated. The aircraft was near aerodynamic stall.
The crew advanced the thrust levers to full power. High-bypass turbofan engines have meaningful spool-up time - thrust response is not instantaneous. Three seconds before impact, a crew member called for a go-around.
The main landing gear struck the three-foot concrete seawall at the approach end of runway 28L. The tail section separated on contact. The fuselage rotated and came to rest off the runway.
Who Was Flying, and Why Does It Matter?
The flying pilot in the left seat had over 9,000 total flight hours and was fully qualified on the aircraft. But he had fewer than 50 hours on the Boeing 777 specifically and was still building familiarity with the 777’s automation behavior. The right seat was occupied by a training captain - a check airman with over 12,000 total hours - acting as pilot monitoring.
Neither pilot was inexperienced by any objective measure. The gap was not total flight hours. It was 777-specific automation familiarity, at a critical phase of flight, after a ten-hour transpacific crossing.
What Did the NTSB Conclude?
The NTSB final report, issued June 2014, identified probable cause as the crew’s failure to monitor airspeed and descent rate, the inadvertent transition of the autothrottle to a non-speed-controlling mode, and what the board specifically labeled automation complacency. Additional findings addressed training practices - particularly manual flying proficiency and autothrottle mode awareness in the 777 transition program.
Boeing issued guidance after the accident with a direct message: when you disconnect the autopilot, do not assume the autothrottle is in the mode you expect. Verify the mode annunciator. Confirm the speed target is active. Never assume system state and stop monitoring.
What Is the Clumsy Automation Paradox?
Earl Wiener, an aviation human factors researcher at the University of Miami, described this dynamic in the 1980s - decades before Asiana 214. He called it the clumsy automation paradox: modern cockpit automation dramatically reduces workload during routine phases of flight, but it concentrates errors in the transitions. Mode changes. Reconfigurations for approach or departure. Go-around execution. These are precisely the moments when automation can be in an unexpected state and crew cognitive load is highest.
Boeing’s own accident analysis identified automation in an unexpected state as a contributing factor across a meaningful percentage of approach and landing accidents. The FAA has issued multiple guidance documents specifically addressing monitoring automation and verifying mode engagement before and during critical flight phases.
Does This Apply Outside the Airlines?
Yes. The mechanism differs by aircraft. The failure mode is identical.
A simple wing-leveler in heading mode will track a heading directly toward terrain. It has no opinion about obstacles ahead. A Garmin GFC 500 in vertical speed mode will hold the commanded vertical speed even as airspeed decays toward a stall - the low-speed awareness system will alert, but the automation will continue executing the original instruction.
The automation does not know what you meant. It knows what you told it. That is as true in a Cessna 172 as it is in a Boeing 777.
Three Practices That Directly Counter Automation Complacency
Verbalize mode states at every transition. Any time you disengage the autopilot, change a mode, or reconfigure for approach or departure, state out loud what the automation is doing. “Autothrottle is in speed mode, target is 137 knots.” The act of verbalizing forces verification rather than assumption. It costs nothing.
Cross-check primary instruments at defined gate altitudes. At the final approach fix, at glideslope intercept, at 500 feet - check airspeed trend, vertical speed, and power setting. These cross-checks exist because automation can silently be in an unintended state, and a structured scan is how you find it before it matters.
Know your automation deeply, not just operationally. Know what your autothrottle or autopilot does when you transition between modes - not just how to use it, but how it behaves. What modes are possible in your specific aircraft? What do the annunciations look like for each? If those questions can’t be answered on the ground, they should not be worked out for the first time at 500 feet on final.
What Has the Industry Changed Since 2013?
Airlines revised initial and recurrent training curricula to emphasize monitoring loops - structured habits for confirming automation state at defined intervals. The distinction between active monitoring and passive monitoring became an explicit training topic in its own right, not just a paragraph in a checklist.
The design question Boeing and the FAA examined - whether an autothrottle should be permitted to enter a passive state without explicit crew action - remains an open debate in aviation engineering. Airbus uses different autothrust logic than Boeing. Neither design is inherently safer; they carry different failure modes and require different training approaches.
Cockpit automation has made aviation measurably safer. Fatal accident rates per departure dropped as glass cockpits became standard. CFIT rates fell after ground proximity warning systems became mandatory. Traffic conflict rates fell after TCAS became standard equipment. The engineering is working.
What it also created is a category of accident that steam gauges could not produce: a system in a state the pilot did not intend, a crew that didn’t know it, and a gap that remained invisible until it was no longer survivable.
The Boeing 777 at San Francisco flew exactly as it was designed to fly. Every system performed within specification until the aircraft met the concrete seawall. The only failure was at the boundary between the crew’s mental model of the automation and its actual state.
That boundary exists in every cockpit with any automation. The only question is how well you know which side of it you’re on.
Key Takeaways
- Asiana Flight 214 crashed on July 6, 2013 at SFO after the autothrottle entered HOLD mode and stopped controlling airspeed during a clear-day visual approach
- The HOLD mode annunciation was displayed on the primary flight display - the system was not malfunctioning; the crew was not monitoring
- Airspeed decayed to approximately 112 knots against a target of 137 knots before the energy state problem was detected
- Three passengers were killed; 187 were injured - every aircraft system performed within design specifications
- Aviation human factors research has documented mode confusion and automation complacency as systemic risks since the 1980s; Asiana 214 is part of a pattern, not an isolated event
- The same failure mode - believing the automation is doing something it is not - applies at every level of cockpit automation, from a basic wing-leveler to a widebody transport
Radio Hangar. Aviation talk, built by pilots. Listen live | More articles