Over 100 of Baidu's Apollo Go robotaxis simultaneously froze in Wuhan traffic Tuesday, trapping passengers inside vehicles that refused to move and causing at least one accident in the resulting chaos. Police confirmed multiple reports of the autonomous vehicles stopping dead in streets and highways across the city, attributing the mass failure to an unspecified "system failure." No injuries were reported, but the incident affected a significant portion of Baidu's 500-vehicle Wuhan fleet.

This isn't just another autonomous vehicle glitch — it's a stark reminder of the single points of failure inherent in centralized AI systems operating physical infrastructure. When your recommendation algorithm crashes, you get bad suggestions. When your robotaxi fleet crashes, you get actual traffic accidents and people trapped in metal boxes. Baidu operates in 26 cities globally and partners with Uber in London and Dubai, meaning this type of simultaneous failure could happen anywhere their systems reach.

What's particularly concerning is how little information Baidu has shared about the root cause. "System failure" tells us nothing about whether this was a cloud connectivity issue, a software bug pushed to the entire fleet simultaneously, or something more fundamental in their autonomous driving stack. The opacity around such a public safety incident raises questions about how much visibility regulators actually have into these systems before approving citywide deployments.

For anyone building AI systems that control physical infrastructure: this is your wake-up call about graceful degradation and fail-safes. If your system can't handle losing connectivity to your servers without trapping humans, you're not ready for production. The robotaxi future might be inevitable, but incidents like this show we're still treating beta software like it's ready for mass deployment.