Waymo Safety Impact (waymo.com)

Waymo published a transparency hub summarizing its rider-only (no human driver) safety data through June 2025 — showing 96 million rider-only miles across Phoenix (46.4M), San Francisco (29.9M), Los Angeles (16.5M) and Austin (3.2M) — and comparing Waymo Driver incident rates to adjusted human-driver benchmarks. The headline findings: large, statistically significant reductions in serious outcomes versus humans — 91% fewer serious-injury-or-worse crashes (0.02 vs 0.23 incidents per million miles), ~80% fewer injury-causing crashes (0.80 vs 3.96 IPMM), and ~79% fewer airbag-deployment crashes (0.35 vs 1.65 IPMM). Reductions are also pronounced for vulnerable road users (pedestrians -92%, cyclists -78%, motorcyclists -89%). Nearly half (47%) of Waymo-reported collisions had a Delta-V under 1 mph, indicating lower-severity impacts when collisions did occur. The report emphasizes rigorous, reproducible methods: comparisons use incidents-per-mile, benchmark adjustments for driving mix and police underreporting (32% adjustment for injury crashes), and linkage to NHTSA SGO reports with downloadable IDs. Waymo notes city-level sample limitations (Atlanta/Austin results not yet statistically robust) and explains differences in reporting thresholds between AV and police data. For the AI/ML community, this provides a rare, data-rich case study of deployed perception/planning stacks reducing real-world harm, while also offering open data and methodology for independent validation and safer-system benchmarking.