An Airbus A320 lost 44 degrees of composure over Dublin Bay. Nothing about it was a flying problem.

On 14 January 2025, an Airbus A320 with 146 people on board was approaching Runway 28L at Dublin. The commander had 7,784 hours, of which 7,623 were on the A320. Within four minutes the aircraft reached a bank angle of 44 degrees, fell below its minimum selectable speed twice, and climbed 700 feet through an air traffic control clearance. Eight minutes later the same pilot flew a clean ILS to an uneventful landing.

That last sentence is the whole point. Whatever went wrong at 14:02 was not a deficiency of skill, because the skill was demonstrably intact at 14:11. The Irish Air Accident Investigation Unit published its findings in Report 2026-008 on 31 July 2026 and classified the occurrence as an aircraft upset. Read as a handling event, it is unremarkable. Read as a study in where attention goes when it is not deliberately directed, it is one of the most instructive reports of the year.

A competent pilot does not stop being competent under load. He stops being able to spend attention on more than one thing.

The arithmetic that closed on paper

Descending through FL135, the crew was given an estimate of 40 track miles to touchdown. The rule of thumb every jet pilot carries — first two digits of the flight level, multiplied by three — returns 39. The numbers agreed, so the question was closed.

The AAIU points out what that heuristic quietly omits: aircraft type, groundspeed, weight, rate of descent, and the time required to decelerate to flap extension speeds. The aircraft was doing 416 knots over the ground with a tailwind. The rule of thumb answers the altitude question and says nothing at all about the energy question, and the energy question was the one that mattered.

This is cognitive miserliness in its purest operational form. The mind reaches for the cheapest available computation, gets an answer inside tolerance, and stops. Nobody in that cockpit was lazy. The heuristic is taught, endorsed and usually correct. It failed here because it was accepted as a conclusion rather than treated as a first approximation requiring a second look.

Having accepted it, the crew then reduced their own margin — requesting a direct routing to MAXEV, the Final Approach Fix itself. The commander, the AAIU records, was comfortable that direct routing to the fix would still permit a stable approach. That belief appears as the first contributory cause.

The energy gap 01,000 2,0003,0004,000 ALTITUDE (ft) 12 NM9 NM 7.1 NM4 NMTHR Published 3.0° glide path MAXEV — cross at 2,500 ft ~1,100 ft high 1,400 ft to lose in 1 NM Actual — manual flight, speed brake full Approach abandoned — orbit requested Paralleling the localiser, south of the inbound course

At 8.1 NM the aircraft was descending through 3,900 ft. To cross the Final Approach Fix at the published 2,500 ft, it needed to shed roughly 1,400 ft inside a single mile. Recovery by ordinary means had already ceased to be available.

Two offers, twice declined

The controller did something that deserves to be noticed. Descending through 4,400 feet, he asked whether the crew was content with the height or wanted extra track miles. When they replied that they intended to go through the localiser, he told them to say the word if they needed further vectors. Two open doors, thirty seconds apart, from a controller who had spotted the trap from the ground.

Neither was taken. The AAIU lists the failure to request extra track miles or entry into a holding fix as a contributory cause.

This is the signature of premature cognitive closure, and it is the same structure that killed forty-seven people at Kegworth. Once a mental model is committed to — we have the miles, we will make this — incoming information is no longer evaluated on its merits. It is filtered for fit. The controller’s question was not heard as evidence that the model was wrong. It was heard as an administrative courtesy that did not apply.

The uncomfortable part is that declining help felt reasonable at the time. It always does. That is precisely why the discipline has to be structural rather than situational: when someone outside the cockpit independently notices your energy state, that observation is data about you, not about them.

Buying drag with attention

Then comes the decision that gives this report its teaching value. The commander disconnected the autopilot. Not because the autopilot had failed, and not to maintain hand-flying currency, but for a single mechanical reason: with the autopilot engaged, the A320 limits speed brake deflection to 25 degrees on spoilers 3 and 4. Disconnect it and you get 40.

Fifteen degrees of spoiler travel, purchased with the entire supervisory layer of the aircraft — in instrument conditions, inside a busy terminal area, at the exact moment the workload was about to peak. The AAIU makes it the second contributory cause: the decision to disconnect and fly manually degraded situational awareness.

This is the inversion that sits at the centre of everything I have written about automation. The autopilot is not a convenience to be surrendered when things get difficult. It is an attention-management device, and its value rises exactly as available attention falls. Airbus says as much in its own guidance — automation exists to relieve the pilot flying of routine handling so that time and resources can go to situational awareness and problem-solving. The rule of thumb is brutally simple: if the only way to make the profile is to switch off the autopilot for extra drag, the profile is already unflyable. That decision is the diagnostic, not the fix.

The first officer put it more elegantly than any textbook when he was interviewed afterwards. Maximising the use of the autoflight systems in scenarios like this one, he said, frees mental capacity — and in freeing capacity, situational awareness increases. He had also, during the event itself, mentioned to the commander that the autopilot was available. The aircraft continued to be flown by hand.

The cone of visibility

What happened next is described, remarkably, by the pilots themselves. The commander told the investigation that while manually flying the approach, his “cone of visibility was reduced”.

Six words. It is rare to get a first-person account of perceptual narrowing this clean in a public safety report, and it is worth more to a training department than any amount of theory. He is describing attention tunnelling as it was experienced: the aperture closing, the periphery dropping away, the world reducing to the one parameter that is screaming loudest.

Both pilots also described startle — the commander mildly, the first officer as intense. We tend to treat startle as an instant, a jolt that passes. It is not. It is a tax on cognition that runs for seconds to tens of seconds afterwards, and the flight data shows what it bought here. At 14:01:39 the bank exceeded 30 degrees for eleven seconds, peaking at 44. At 14:02:29 the airspeed fell below the minimum selectable speed. At 14:02:35 the autothrust was switched off. Three separate degradations inside fifty-six seconds, each one narrowing the aperture that produced the next.

The one thing that went right in that window went right because someone outside the tunnel was watching. The first officer called the bank. The commander corrected it immediately. Monitoring is not a passive state of being in the other seat — it is an active skill, and it did its job here.

The flight director nobody followed

Here is the detail that should be on a slide in every recurrent course. As the aircraft turned east, it was in OP DES with idle thrust commanded and the flight director bars engaged. The flight director was therefore giving a pitch-down command to hold the target speed of 139 knots.

The command was not followed. The speed decayed anyway.

The information was present, it was correct, and it was directly in front of him on the primary flight display. It was simply unattended. This is what makes the event an attention failure rather than a knowledge failure — the guidance was never in dispute, it was never looked at.

Then the second-order problem. When the speed dropped below the minimum selectable speed, the autoflight speed protection did exactly what it is designed to do: the flight director bars were removed for about a second and the autothrust reverted to speed mode to recover the target. The commander read this as evidence that the autothrust was malfunctioning. He selected TOGA, switched the autothrust off, and it stayed off for the remainder of the flight.

Nothing had failed. The aircraft was protecting itself and announcing that it was doing so. But under a closing aperture, an unexpected annunciation is far more likely to be interpreted as a fault than as a designed response — and the cost of that misreading was the last remaining layer of automation, discarded at the point of maximum need. Attention control is not only knowing where to look. It is holding the mode model that makes what you see mean the right thing.

When the pilot flying does the monitor’s job

Alarmed by the speed decay, the commander selected Conf 3 and then Conf FULL himself — a task normally called for by the pilot flying and executed by the pilot monitoring. The first officer, interviewed later, explained it as a consequence of how dynamic the situation had become.

The AAIU is direct about the consequence: when the pilot flying also performs the functions of the pilot monitoring, crew resource management degrades and the other pilot can be left feeling uninvolved in the operation of the flight.

I would put it more sharply. Self-selected flaps are a visible external marker of saturation, and they invert exactly the response that saturation demands. Under load the human instinct is to reclaim control, not to delegate it — to shrink the circle rather than widen it. The result is a two-pilot crew that has quietly become one pilot and a spectator, at the moment it most needs to be two.

Which raises the question the first officer answered honestly. Asked whether he considered taking control, he said he thought about it briefly during the high bank angle, but the commander responded correctly to his call, so he judged it unnecessary. That is a defensible judgement. It is also the judgement every monitoring pilot makes under uncertainty, usually alone and in about two seconds. The operator afterwards reissued its guidance on the protocols for taking control — which is the correct systemic answer, because the decision should not rest on an individual’s improvised threshold.

The fatigue the model could not see

The operator’s fatigue software returned a peak score of 47 against an operational threshold of 70. Comfortably compliant, and the AAIU records it as within the acceptable range.

Beside that number sit three facts from the same report. The commander preferred late duties. His roster had rotated him from lates to earlies across a single blank day. And he told the investigation that he had not slept well the night before and may have been tired.

The AAIU makes the crucial observation itself: the score reflects likely sleep opportunity, not sleep actually obtained or its quality. The model measured the roster. It did not, and cannot, measure the man.

I am not arguing that fatigue caused this event; the report does not, and neither will I. I am arguing something narrower and more useful. Attention is the first faculty degraded by poor sleep, and the metric we use to govern it is blind to the individual sitting in the seat. A biomathematical score is a rostering tool. It was never a fitness-for-duty test, and treating a green number as a statement about a particular pilot on a particular morning is a category error we make routinely.

Sixty-three days

The occurrence happened on 14 January 2025. It was reported voluntarily on 18 March — sixty-three days later. By then both the cockpit voice recorder and the flight data recorder had been overwritten. The crew could not be interviewed until roughly three months after the event, and the AAIU notes that any inconsistencies between recollection and recorded fact can reasonably be attributed to elapsed time.

Everything above — the cone of visibility, the startle, the misread annunciation — survives only because two pilots were unusually candid three months after the fact, and because the operator’s flight data monitoring system, radar traces and ATC recordings filled the gap the recorders should have filled.

We talk about reporting culture almost entirely in terms of willingness. This report is a reminder that latency matters just as much. A report filed on the day preserves the recorders. A report filed two months later preserves only memory. Both are voluntary; only one is evidential.

What held

It would be a poor reading of this report to take only the failures from it. Four things worked, and each is a defence worth funding:

The monitor monitored. The first officer held his situational awareness throughout and called a deviation the pilot flying had not perceived.

The envelope held. Speed protection activated correctly on both excursions. The aircraft never left the normal flight envelope.

The controller intervened. He identified the developing energy problem from the ground and offered a way out twice, unprompted.

The crew told on themselves. No alert, no damage, no obligation. The occurrence entered the safety system because two pilots chose to put it there.

The operator’s response was equally sound. Following an internal review, it issued a Flight Crew Instruction in December 2025 prohibiting visual orbits outright and setting out the only acceptable alternatives: entry to a published holding fix, additional track miles from ATC, or — stated as the preferred option wherever there is any doubt about the outcome of an approach — a go-around. The instruction also names, explicitly, the trap this crew fell into: getting high or fast during the initial approach may itself be the product of an earlier loss of situational awareness, and attempting to recover it with an orbit may degrade the situation further.

Because that action had already been taken, the AAIU sustained no safety recommendations. That is what a functioning system looks like.

What to take from it

Nobody was hurt. Nothing was bent. There are no recommendations to point at. By every conventional measure, this is a minor report about a routine day that went briefly untidy over Dublin Bay.

Which is exactly why it should be taught. The catastrophic reports are studied because they are dramatic; the instructive reports are the ones where nothing was destroyed except a margin. Here we have a highly experienced pilot, a serviceable aeroplane, an alert first officer, a helpful controller, benign weather, no time pressure, and a fatigue score well inside limits — and four minutes in which a routine approach came apart, entirely inside one man’s head.

The failures were sequential and every one of them was about where attention went: a heuristic accepted instead of interrogated; help declined because the model had already closed; automation traded for drag; a flight director command unattended; a protection mistaken for a fault; a monitor’s task absorbed by a saturated pilot. Not one of them required a lapse of skill.

We train handling relentlessly, because handling is visible and measurable. Attention we largely leave to itself, and then express surprise when a 7,600-hour pilot’s world contracts to a cone. The commander of EI-DEE gave us the language for it in six words during an interview he did not have to give. We should use them.

Source

Air Accident Investigation Unit Ireland, Final Report 2026-008: Incident involving an Airbus A320, registration EI-DEE, on approach to Dublin Airport, 14 January 2025. Published 31 July 2026. Investigator-in-Charge: Ray Jordan.

The AAIU report is the authoritative account. In accordance with Annex 13, EU Regulation 996/2010 and SI 460 of 2009, safety investigations are not concerned with apportioning blame or liability, and this commentary is offered solely for training and safety-learning purposes.

Capt. Amit Singh, FRAeS

Founder of the Safety Matters Foundation and author of mindFly: Follies, Realities & Human Factors. An airline captain with over 18,000 hours and 35 years in civil aviation, he writes and lectures on attention management, cognitive traps and crew performance.