Limitations of Robotaxi Safety Claims with Phil Koopman

A deep dive into why “AVs are safer than humans” headlines are shaky: safety studies often ignore changing software, remote-operator roles, empty-vehicle miles that reduce “injury crashes,” selective reporting, weak human-driver baselines (especially COVID-era data), and risk exported to other road users when vehicles stop or phantom brake. The big warning: past performance doesn’t prove safety at massive scale or in new cities with different conditions, and rare fatalities need far more miles to measure. Skepticism also targets closed-door industry “standards” shaping federal rules, plus quick hits on messy recall fixes, rear-camera failures, and an investigation into Comma.ai after crashes and fatalities.

https://avsafety.substack.com/p/a-guide-to-the-limitations-of-robotaxihttps://www.yahoo.com/news/us/articles/nhtsa-wants-ideas-fix-recalls-120949573.htmlhttps://static.nhtsa.gov/odi/rcl/2026/RCLRPT-26V605-9379.pdfhttps://static.nhtsa.gov/odi/rcl/2026/RCLRPT-26V597-1248.pdfhttps://static.nhtsa.gov/odi/rcl/2026/RCLRPT-26V598-8506.pdfhttps://static.nhtsa.gov/odi/inv/2026/INOA-PE26007-17274.pdf