Move Fast and Break Nothing (www.theatlantic.com)

An independent review of Waymo’s disclosed crash reports finds the majority of its serious incidents weren’t caused by its self-driving stack but by human factors and other road users—most notably repeated cases where passengers flung open doors and struck cyclists or scooter riders. Waymo touts a strong safety record across roughly 96 million fully autonomous miles, claiming 91% fewer accidents causing serious injury than an average human driver over the same distance. Experts note caveats—Waymo’s fleet is new and carefully maintained—but agree the company’s cautious, data-heavy approach has meaningfully reduced tech-caused harm so far. The story matters because it contrasts two AI deployment cultures: Waymo’s slow, iterative, zone-limited expansion (five cities, strict highway/rider rules, sensor-cleaning measures) versus the “move fast” posture seen in other AI and robotaxi rollouts (e.g., Tesla, Cruise, and rapid chatbot launches) that have produced high-profile failures and regulatory scrutiny. Technically, the piece underscores the hard reality of edge cases—weather, unexpected human interaction, high-speed highway scenarios—and the enormous dataset, engineering rigor, and capital (Waymo’s 16-year buildup and Alphabet losses) required to mitigate them. The upshot: deliberate, constrained deployment and long-tail data collection may be the most viable path for safe scaling of embodied AI systems, but a single software-linked fatality could quickly reshape regulator and public trust.