Miami Amazon Air Crash: Analysis or Hindsight?
On 6 September 2026, 21 Air Flight 7598, a Boeing 767-300 freighter operating on behalf of Amazon, arrived at Miami International Airport following a flight from San Juan, Puerto Rico. The aircraft crossed the threshold of Runway 30 at high speed, made its first runway contact about 3,750 feet beyond the threshold and ultimately departed the end of the runway at around 113 knots. It continued beyond the airport boundary and struck two vehicles. Five people on the ground were killed, three were seriously injured, and the two pilots received minor injuries. The aircraft was substantially damaged and a post-crash fire developed in the right wing.
There is no question that the approach and landing described in the NTSB's preliminary report raise some serious questions. The aircraft was fast, it was not fully configured at points where we would normally expect it to be, the first officer repeatedly commented on the excessive speed, and the landing occurred a long way down the runway. The tower controller also did not provide the crew with the required wind information when issuing the landing clearance, despite a significant wind shift having occurred. What we do not yet have is a probable cause, an analysis of why the crew behaved as they did, or a complete understanding of the organisational and operational context surrounding the accident. The NTSB has been very clear that the information currently available is preliminary and subject to change.
So why I am I writing this blog? Well, I recently watched a YouTube analysis of the preliminary report. There is plenty in the video that is technically interesting, and his knowledge of the aircraft is obvious, but I found myself becoming increasingly uncomfortable (frustrated perhaps) with the way the story was being told. By the end, the crew's actions had been described as "incredible incompetence", the first officer had effectively been admonished to "speak up" because "five people died", and a running count had been constructed of somewhere between 11 and 14 occasions on which the pilots supposedly ignored a mandatory requirement to go around.
Perhaps the final investigation will conclude that serious individual performance failures occurred. It may even support some of that criticism. My concern is that we don't know that yet, and turning a preliminary factual report into a morality tale about what competent pilots 'obviously' should have done risks teaching us far less than the accident potentially can.
Eleven missed go-arounds sounds compelling. Is it useful?
One of the dominant themes in the video is the number of occasions on which the crew supposedly should have gone around. The 1,000-foot call, sink-rate warnings, the 500-foot gate, further sink-rate warnings, terrain alerts, configuration, touchdown position and other parameters are progressively counted on the presenter's fingers until the number reaches double figures. Each becomes another piece of evidence against the crew, and when viewed from outside the aircraft, with the flight data plotted in front of us and knowledge of what happens next, the argument feels quite irresistible.
There is another possibility worth considering, however. A warning being generated is not the same thing as a warning being cognitively processed. The fact that a sound was present on the flight deck does not necessarily mean it entered the pilots' conscious awareness in the way it enters ours when we listen to the recording afterwards. Under high workload and intense attentional focus, auditory information can be significantly degraded or excluded from conscious processing. Anyone who has tried to speak to somebody completely absorbed in a difficult task will recognise the phenomenon; the sound reaches their ears, but in a meaningful sense they do not "hear" you.
Look at the situation from inside the cockpit rather than from our comfortable position afterwards. The aircraft is fast, the captain is trying to configure it, the approach is deteriorating, the aeroplane requires increasing attention and the amount of spare cognitive capacity available to recognise that deterioration is probably shrinking at precisely the moment it is needed most. Each additional problem potentially increases workload, which further reduces the capacity to recognise the next problem. Counting every alert retrospectively as another independent conscious decision to ignore a mandatory go-around may therefore create an illusion of repeated decision-making where there may actually have been one developing state of attentional capture.
Don't get me wrong, this doesn't make continuing the approach appropriate. It does however make the behaviour something worth understanding.
What made continuing make sense?
We know quite a lot about the final few minutes of the flight, but we currently know very little about the environment that shaped them.
What was the company's attitude towards go-arounds?
How frequently were they performed?
Were crews supported when they made conservative decisions, or did a go-around generate questions about fuel, cost and operational performance?
Had either pilot previously been challenged for making a decision that caused delay or additional expense?
Was there an informal expectation that experienced crews should be able to "make it work"?
None of those things may have existed, and it would be just as wrong for me to assert that they did as it is to assert that they played no part. The point is that these are exactly the sorts of questions a (Human Factors) investigation should explore before deciding that what we're seeing is simply incompetence.
There is another question I would like to ask: Had these pilots, or other pilots within the organisation, recovered unstable approaches before? If an aircraft has previously arrived high or fast, been brought back within parameters and landed safely, what lesson was taken from the experience? If that happened repeatedly without consequence, did the boundary between recoverable and unacceptable gradually move?
We recently wrote about this in relation to the Challenger Space Shuttle disaster. Diane Vaughan called the phenomenon normalisation of deviance. An abnormal condition doesn't necessarily become normal because people stop caring; sometimes it becomes normal because experience repeatedly tells them that they can manage it. Successful outcomes then reinforce the behaviour until the deviation itself stops feeling particularly unusual. We have no evidence yet that this happened at 21 Air, but it is a much more useful line of enquiry than simply asking why professional pilots failed to do what the rulebook told them to do.
"Speak up" isn't enough
The preliminary evidence tells us that the first officer cautioned that the aircraft was too fast and made additional comments about excessive speed, while the captain did not consistently respond. The video interprets this principally as a failure of assertiveness, telling first officers that they are "not a potted plant" and that after repeated unsuccessful calls the first officer should ultimately have taken control and initiated the go-around.
There is some validity in the underlying point. Effective Crew Resource Management requires communication to escalate when an important safety concern isn't acknowledged. Models such as PACE (Probe, Alert, Challenge, Emergency) exist precisely because politely mentioning a concern once may not be enough. If "we're too fast" doesn't change anything, the communication needs to become increasingly explicit until there is no ambiguity about the problem or the required action.
Where I differ from the YouTube analysis is in assuming that because the words were spoken, they were necessarily received and understood. If the captain was experiencing high workload or attentional tunnelling, the first officer's comments may simply not have penetrated.
Closed-loop communication and escalating assertiveness are important, really important. Communication isn't complete when information leaves my mouth; it's complete when I have evidence that you have received and understood it. The absence of a meaningful response should therefore have been information in itself, but rather than stopping at "the first officer should have been more assertive", I'd want to know what the crew had been trained to do when normal challenge-and-response communication stopped working, how those behaviours were practised and whether the authority gradient or organisational culture made escalation genuinely achievable.
Procedures are there for a reason, but professional judgement has a place
There is a broader issue here that goes beyond this accident. Stabilised approach criteria are extremely valuable because they provide crews with predetermined boundaries at a time when continuing can become increasingly seductive. They reduce ambiguity and help prevent continuation bias. If an operator's procedure says an unstable approach at a particular gate requires a go-around, then that requirement is there for a reason, and nothing I'm saying here should be interpreted as an argument for casually disregarding it.
What I resist is the idea that good airmanship can be reduced to mechanically applying a rule regardless of context. Professional pilots aren't employed merely to execute procedures; they're employed to manage risk and achieve safe outcomes in a complex, often dynamic, environment where circumstances occasionally fall outside what a procedure anticipated. Imagine an aircraft at 1,000 feet in benign weather, marginally outside a stabilised criterion, carrying a passenger whose medical condition is rapidly deteriorating. If the crew can readily correct the deviation and safely land the aircraft, I would be very uncomfortable with an analysis that treated a go-around as unquestionably safer simply because a line in a manual had been crossed. A go-around carries risks too, and in that situation delay itself may carry a very real human cost.
The important distinction here is between routine non-compliance and professional judgement. We should be extremely cautious about using judgement as an excuse for habitual deviation, but equally cautious about creating systems in which compliance becomes the objective rather than the safe outcome the procedure was designed to support.
The advantage of knowing the ending
Perhaps the biggest problem with analyses like this isn't that the technical observations are wrong. Many of them may ultimately prove entirely correct. The problem is that hindsight changes the story.
Watching the approach after five people have died, we know exactly which pieces of information matter. Every speed call becomes significant. Every alert sounds obvious. Every opportunity to go around appears enormous. We can pause the video, count the warnings, examine the data and ask how anybody could possibly have continued.
The pilots had none of those advantages. They experienced the event forwards, once, in real time, without knowing the ending.
That doesn't absolve them of responsibility, nor does it mean there should never be consequences when professional standards aren't met. It simply means that if our analysis begins and ends with what they should have done, we're unlikely to understand why they did what they actually did.
The NTSB investigation has a great deal of work still to do. It will examine operations, Human Performance, air traffic control, meteorology, aircraft performance and other aspects of the accident before reaching its conclusions; the preliminary report itself expressly does not provide probable cause.
The final investigation will be much more enlightening than counting mistakes from the comfort of hindsight.
The useful question isn't "How could they possibly have ignored eleven reasons to go around?"
It's "What was happening in that cockpit, and around those pilots, that made continuing seem like the thing to do?"
Until we understand that, we're analysing the outcome. We're not yet learning from it.
Mike Mason and Sam Gladman help leaders and teams perform under pressure. Drawing on decades of experience in military aviation, Human Factors, accident investigation and organisational learning, they work with safety-critical organisations to improve decision-making, teamwork and leadership when conditions are uncertain. Through On Target Teaming, they combine practical tools, immersive simulations and real-world case studies to help teams perform at their best when it matters most.
If you'd like to learn more about how On Target can help your team, contact Mike and Sam at info@ontargetteaming.com.
Comments