For more than a year prior to a fatal crash in Arizona, Uber’s self-driving cars failed more often, and more dramatically, than competitors’ autonomous vehicles. At the same time, Uber reduced some safety precautions, and was sometimes misleading in its description of its program and its failures. And regulators in Arizona, the locus of Uber’s testing, have taken little action to protect residents despite those worrying signals.
It is not yet clear whether Uber’s system was at fault or not in the latest crash, but Uber has now halted all testing of its autonomous vehicles, with no clear timeline for reactivation.
The New York Times reported yesterday that, in October of last year, Uber altered its testing program by putting only one safety monitor in each autonomous car rather than two, over the safety concerns of some employees. That move came despite evidence of deep problems with Uber’s autonomous vehicle efforts, dating back as far as December of 2016. That’s when Uber vehicles were seen running red lights in San Francisco. The company first blamed one of its human safety drivers, before it was uncovered in February that the problem was actually with the autonomous system itself.
Evidence quickly emerged that this was not a freak occurrence. In March of 2017, Recode obtained internal documents showing that human drivers had to take over from Uber’s system very frequently relative to the same numbers for other self-driving efforts. Then, the same month, a self-driving Uber flipped on its side in Arizona, though Tempe police found the Uber was not at fault.
Get Data Sheet, Fortune’s technology newsletter.
These public troubles seemed to reflect internal problems. The leadership of Uber’s self-driving car unit has frequently been described as troubled, with high levels of engineer attrition. Meanwhile, Google spinoff Waymo alleged in an explosive lawsuit that Uber had stolen technology from it by way of former Waymo executive Anthony Levandowski, who was fired from Uber in May of last year.
Finally, just a few days before this week’s fatal crash, an Uber vehicle in self-driving mode crashed into another vehicle in Pittsburgh. Fault in that crash had not been determined as of recent reporting.
San Francisco regulators put a quick stop to Uber’s testing there in the wake of the red-light incident. But even after sustained warning signs, Arizona officials took no such action, and reiterated this week that there were no plans to change the state’s hands-off regulatory approach.
Many observers believe that the future of Uber hinges on the success of its autonomous driving program. The company regularly posts quarterly losses with few historical parallels, even as regulators and critics argue with growing vehemence that the company is exploiting and underpaying its drivers.
Autonomous vehicles were intended to square that financial circle by taking driver pay out of the equation. The company, according to the Times, had planned to launch a self-driving car service in Arizona by December. CEO Dara Khosrowshahi has canceled a planned April visit to Phoenix to check in on the program’s progress, though the company claims that change is unrelated to the crash. The company’s bigger plans could also now wind up delayed – including not only progress on the road to autonomous driving, but towards its hotly anticipated initial public offering.
