Flight 7598 crossed a public road and hit a van with seven people inside. The vetting question beneath a contracted cargo flight is older than the aircraft, and it doesn’t go away when the branding does.

On September 6, 2026, a Boeing 767 wearing Amazon colors ran off the end of Runway 30 at Miami International Airport at 1:53 p.m., crossed the airport fence and a public road, and struck a van with seven people inside before catching fire. Five people died and five were injured, Miami-Dade authorities said. The FAA identified it as 21 Air Flight 7598, arriving from Luis Muñoz Marín International Airport in San Juan, Puerto Rico. The plane on the news was an Amazon plane. The airplane in the federal file was not. The FAA identified the flight as 21 Air Flight 7598, and the operating carrier, not the customer whose logo is on the tail, is the one that holds the certificate under which it flew. That distinction is where this story actually lives, and it is the distinction that gets lost in every headline that starts with the word Amazon.

Amazon arranges cargo capacity through airline partners; the operating airline flies under its own air carrier certificate, and the FAA issues and oversees that authority. In plain terms, Amazon books the space and somebody else flies the metal. The industry runs this as an ACMI arrangement: the operator supplies the aircraft, crew, maintenance, and insurance. 21 Air, LLC is an all-cargo airline headquartered in Greensboro, North Carolina, operating a fleet of Boeing 757s and 767s. It began flying for Amazon Air in late 2024, and this particular airframe started operating for Amazon under 21 Air in January 2025. When the NTSB reconstructs this crash, it examines the certificate holder, 21 Air. The logo does not carry a certificate. The company underneath it does.

Chairwoman Jennifer Homendy took a go-team to Miami and described the scene as “utter devastation,” with debris from both the passenger vehicles and the aircraft across the site. The aircraft was a 32-year-old Boeing 767-33AER converted freighter, registered N1997A, delivered new to Belgian carrier Sobelair in 1994. The maintenance history is where former DOT Inspector General Mary Schiavo told reporters the NTSB will spend real time, and the reason is the ownership chain. This tail spent its passenger life across three continents before it ever hauled a box.

The airframe went from Sobelair to Vietnam Airlines in 2000, to Air Europa in 2003, to Kenya Airways in 2006, to Russian carrier Nordwind Airlines as VQ-BRA in 2010, then was withdrawn from service and stored at Shannon, Ireland in 2014. It was ferried to Tel Aviv, converted from a passenger jet to a Bedek Special Freighter by Israel Aerospace Industries between November 2015 and June 2016, and joined Atlas Air as N1997A. The Russian operating period is not a scandal by itself. It is a records question. Every registry change is a handoff of the maintenance file, and every handoff is a place where the paper trail can thin out. Whether any of it touched this landing is for the NTSB to determine, and it has not.

Runway 30 at Miami is more than 9,000 feet long, and Flightradar24 tracking data showed the jet rose above the normal approach path before dropping back toward position; the plane had passed through a developing thunderstorm on approach. ADS-B data received by Flightradar24 showed the aircraft’s speed at 112 knots as it exited the usable runway, with the last received speed at 69 knots. A jet still doing 112 knots where the pavement ends is a jet that touched down long, and aviation safety experts told reporters the touchdown zone is the first 3,000 feet. Miami had no engineered material to stop what happened next.

The technical term for that stopping material is EMAS, an engineered materials arresting system, and the plain version is a bed of crushable blocks at the end of a runway that swallows the landing gear like mud and drags a plane to a stop. The FAA began this research in the 1990s, and EMAS uses crushable material at the runway end so the tires sink in and the aircraft rapidly decelerates. A standard EMAS brings the critical aircraft to a stop when it enters at 70 knots or less, governed by FAA Advisory Circular 150/5220-22. The rule behind it all is the runway safety area. The FAA requires a 1,000-foot safety area at the end of a runway, and those standard areas have historically been difficult to obtain at airports built before the standard existed. Miami had the buffer. Miami did not have the bed.

This is where the compliance question and the safety question stop being the same question. Miami’s airport appears to be in compliance and wasn’t required to have an arresting system because it has a 1,000-foot safety buffer at the end of its runways, said Mike O’Donnell, the FAA’s former director of airport safety and accident investigations. Compliant is a status. It tells you the airport met the standard. It does not tell you the standard would have stopped this airplane, and the crushable bed the FAA credits with saving lives elsewhere was not there to try. The requirement that drove EMAS adoption traces back to safety-area shortfalls at older airports, and it matters here that a 1,000-foot buffer and an EMAS bed are not the same protection, even when the regulation treats the buffer as enough.

Homendy said the plane hit both a van with seven people inside and an SUV; the van was inside the airport perimeter and the SUV was outside it. The people who died were not on the airplane. They were on a road that warehouse workers use, next to a busy airport in a dense city. The liability exposure here is not one company’s. The customer that arranged the capacity, the certificate holder that flew it, the owner of the airframe, the airport authority that chose its safety configuration, and the maintenance contractors across a 32-year life are all now inside an NTSB reconstruction that will take a year or more. Every one of those relationships is a place where somebody decided the arrangement was good enough, and a jury will eventually be asked whether it was.

The 2019 crash of Atlas Air Flight 3591 also involved a Boeing 767 operating for Amazon Air. That one went down on approach to Houston and killed three. This one overran in Miami and killed five on the ground. Two different carriers, two different failure modes, the same customer’s boxes in the hold. That is not a finding against Amazon. It is a pattern worth naming while the investigation runs, because the contracted-capacity model puts the customer’s brand on aircraft the customer does not fly or maintain, and the public reads the brand as the operator every time.

A 21 Air 767 overran Runway 30 at 112 knots, crossed a public road, hit two vehicles, and killed five people, and the NTSB has a 32-person team documenting the site. What the record does not establish is why the airplane touched down where it did, whether weather, a tailwind, the aircraft’s mechanical condition, the crew’s decision not to go around, or some combination put it there, and none of that is known until the NTSB says it is. If you operate, contract, or insure aircraft in a branded-capacity arrangement, the directive is the same one that survives every one of these: know which entity holds the certificate on every flight your name rides on, pull the maintenance file on the specific airframe and not the fleet average, and treat the airport’s safety configuration as a variable you selected, not a given you inherited.