Three crashes in the first day:Tesla in Austin (arstechnica.com)

Tesla’s freshly opened autonomous testing in Austin ran into trouble almost immediately: company crash reports spotted by Brad Templeton show three crashes on the first day of testing (July 1), plus a fourth parking-lot incident that went unreported. Two crashes were rear-end collisions caused by other vehicles, while a third involved a Model Y—with a required safety operator aboard—striking a stationary object at low speed and producing a minor injury. Tesla has redacted many details in the reports, and CEO Elon Musk disclosed that the company had only logged about 7,000 miles of testing by the time of the July earnings call. For the AI/ML community this is significant on multiple fronts: it highlights the risks of limited real-world validation, the limits of “first-mile” deployments in permissive regulatory environments, and concerns about transparency in incident reporting. By contrast, legacy players such as Waymo report crash rates more than two orders of magnitude lower (60 crashes over 50M miles, and now over 96M miles driven), underscoring the value of extensive mileage and conservative validation for safety-critical perception and planning systems. The mix of at-fault and not-at-fault collisions, the presence of a safety operator, and heavily redacted reports raise questions about dataset completeness, reproducibility of failure analyses, and the need for clearer public metrics on autonomous system safety.