Atlas Air 3591—Incompetence at the Controls

Atlas Air flight 3591, a Boeing 767 freighter operated for Amazon’s Prime Air fleet, crashed into Trinity Bay during an approach into Houston, Texas on 2019-02-23, killing the captain, first officer, and a Mesa Airlines captain riding as a passenger in the jumpseat. This was the first fatal crash of a Boeing 767 freighter. The plane struck the ground in a steep descent well above its maximum certified speed.

Investigation found the cause to be the first officer’s inputting extreme nose-down commands following inadvertent operation of the go-around switch, probably in reaction to a somatogravic illusion caused by the acceleration due to the go-around activation and failure to use the flight instruments to determine the actual attitude of the aircraft.

Wikipedia notes:

First officer [Conrad Jules] Aska had also experienced training difficulties with Atlas Air, more so than [captain] Blakely. He had also recorded training failures with previous employers. Another Atlas Air 767 captain who had flown with Aska described him as a “nice guy” and “definitely in the top half of the people I’ve flown with,” though he did not state any issues regarding training. Aska’s first issues were reported in July 2017, the same month he joined Atlas Air, when he was declined an oral exam for his type rating on the 767 as he needed remediation training. Following the training, he passed the oral exam. Aska then went through five fixed-base (non-moving) flight simulator sessions, experiencing difficulties with normal procedures, and underwent more remediation training. In August, following two full-flight simulator training sessions, Aska’s simulator partner complained that he was being “held back.” Atlas Air ultimately had to restart full-flight simulator training for Aska because no other pilots remained in his training class to partner with him.

Aska’s first checkride on the aircraft ended in failure due to poor crew resource management (CRM) and improper aircraft control. His examiner described him as stressed and lacking situational awareness. Aska underwent remedial training on September 25 and the next day, he reattempted his checkride successfully, receiving his type rating on the aircraft.

Investigators concluded that Aska had deliberately concealed his spotty training record when he interviewed with Atlas Air, taking advantage of shortcomings in the FAA pilot records database, which was criticized by the NTSB.

First officer Conrad Jules Aska was a citizen of Antigua. Time reports, “Pilot in Amazon Cargo Plane Crash Repeatedly Flunked Flight Tests”:

Aska didn’t mention in his job application to Atlas that he’d been employed briefly by two other carriers, but had departed after being unable to complete training, according to the NTSB.

The airline wouldn’t have hired him if it had known his complete record, the director of training told investigators.

See if you can guess the words which never appeared in either the official investigation nor media coverage of this accident.


The issue of audible warnings in the cockpit is complex. At times, they could prevent disaster, other times, they may cause confusion/sensory overload and contribute to a crash. In many of the crashes we have seen here, crash analysis mentions how the aircraft’s status (e.g. autopilot on/off, flight director mode, altitude hold, heading mode, etc, was silently displayed on the HSI, but went unnoticed - as here.

Now, I think a case could be made that auditory announcement of something as critical as TOGA power initiation should be made whenever initiated. Even if not every time that switch is actuated, isn’t it true that the flight control computer could be programmed to “know” when such activation was likely inadvertent, given the overall flight situation? Expanding on this concept, might not annunciation of significant changes in control settings be made subject to the superior “situational awareness” of the flight control computer, when it comes to auditory alert? As well, we have seen crashes result when one pilot made a critical change which was not announced to the other pilot. BTW, I do not mean announcement via bells and whistles. Rather, a spoken “TOGA power” announcement, for example. I haven’t enough real world experience in complex cockpits to judge the practical utility of this. The thought is prompted by the rapid advancement of AI capabilities, which could make a difference given the number of variables and situations in flying complex modern aircraft.

As mentioned in another thread today, pilots need to know for certain, e.g. whether (s)he is flying the plane, the computer is, or which partial-component of flight automation is operant. Lack of this awareness has also resulted in disasters.


Abandonment of merit in favor of our betters’ applications of “equity” in hiring is likely providing daily examples of actual results. Witness the recent increase in runway incursions and near-misses. Alas, we are headed for a “new normal” of lots more costly mishaps and preventable deaths. Words and ideas will be omitted, as will root cause analyses. Calls for “better training” will resound. Individual agency and responsibility are things of the past.


It is interesting that the one field of endeavor in which merit still reigns supreme is – professional sports. No US professional sports team “looks like America”. European soccer teams are rapidly heading in the direction of the “All Blacks” – with apologies to New Zealand’s rugby squad. And yet there is not a squeak of protest about discriminatory results from the Usual Suspects.

Presumably, the key difference is that the impact of merit (or its lack) is immediately and unequivocally and publicly obvious after a mere 90-minute game. Of course, crashing a plane is unfortunately also rather obvious.