Tempe, AZ police release video of moments before fatal Uber autonomous car collision; safety driver repeatedly looks at something off-camera before collision
Here's what the car — and its operator — saw. — Three days after an Uber self-driving vehicle fatally crashed into a pedestrian in Tempe …
Context & Ripple Effects
Two days after the Tempe crash, Tempe police had told reporters the collision was likely unavoidable in any mode, pointing blame away from Uber's system preliminary investigation. This newly released dashcam footage complicates that framing: it shows the safety driver repeatedly looking at something off-camera in the seconds before impact.
The video lands alongside independent analysis arguing the car's own onboard sensors should have detected the pedestrian well before contact — meaning the record now points to failures at two layers, the automated perception stack and the human monitoring it. That dual-failure picture is what made this case the reference point for how autonomous-vehicle oversight gets judged.
First-order effects
- Public scrutiny shifts from the software to the safety driver: the footage makes inattention visible in a way that contradicts the 'unavoidable' narrative Tempe police offered after their preliminary review.
- Uber faces immediate pressure over its human-in-the-loop design, since the same recording that indicts the driver also documents that the vehicle did not brake for a pedestrian its sensors should have registered.
Second-order effects
- Police follow-up reporting later found the driver was streaming a Hulu show during the drive per police findings, hardening the case that the monitoring role was unenforceable as staffed — a problem every AV operator running safety drivers now has to answer for.
- Regulators and insurers get a template for assigning fault between operator and human monitor, since the sensor analysis shows neither layer alone would have prevented the crash.
Third-order effects
- The accountability question outlives the news cycle: by 2020 the safety driver was charged with negligent homicide charged with negligent homicide, establishing that individual monitors — not just companies — carry criminal exposure when automation depends on their vigilance.
- If the pattern holds, the industry's reliance on attentive human fallback drivers becomes structurally untenable, pushing operators toward either removing the human role entirely or building auditable systems that do not assume one.
The trend: Autonomous-vehicle incidents are being adjudicated less as engineering failures than as tests of whether human oversight roles are legally and practically enforceable.