Cruise self-driving car injures multiple people in San Francisco collision
A Cruise robotaxi collided with a Toyota Prius during a left turn in San Francisco on June 3, 2022. The accident caused multiple injuries and prompted a federal probe into the company's autonomous driving software.
The autonomous system stopped mid-turn in the intersection, leading to a collision with an oncoming vehicle.
Key facts
- What
- A Cruise robotaxi collided with a Toyota Prius during a left turn in San Francisco on June 3, 2022.
- Incident date
- Jun 3, 2022
- Who
- Cruise
- Failure mode
- Agentic Action Error
- AI surface
- Autonomous System
- Severity
- High
What happened
On June 3, 2022, a Cruise autonomous vehicle collided with a Toyota Prius while performing a left turn at a San Francisco intersection. The collision resulted in injuries to occupants in both vehicles. The incident led to an investigation by U.S. federal agencies.
What broke inside the model
- 01 · TriggerAn agent plans a multi-step task.
- 02 · Model stepIt chooses a wrong or destructive action.
- 03 · Control gapNo confirmation gate guards the write.
- 04 · FailureThe action commits to a system of record.
- 05 · ConsequenceData is changed or destroyed irreversibly.
A wrong action commits, and the step is written before anything can stop it.
The system failed to safely execute a left turn in the presence of oncoming traffic. It stopped the vehicle before the turn was complete, obstructing the intersection and causing the oncoming Prius to collide with it.
What it cost
Sources
- PressGM's Cruise Self-Driving Car Crash Injured Multiple Peopleinsideevs.com
- PressU.S. agency probing self-driving Cruise car crash in Californiareuters.com
Cite this entry
https://failureindex.ai/failures/cruise-self-driving-car-injures-multipleAI Failure Index. "Cruise self-driving car injures multiple people in San Francisco collision" (FI-0692). Realm Labs. https://failureindex.ai/failures/cruise-self-driving-car-injures-multiple (indexed Jun 22, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0692. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AgentRealm
Realm can sit inline on the agent's action path and require that a destructive or high-consequence action clears a real check before it executes, so 'delete and recreate' or a wrong write is stopped at the moment of intent, not explained in the post-mortem.