11
Waymo’s Co-CEO Warns AI Builders: “There Is No Undo Button”
Waymo’s Co-CEO Warns: "Move Fast and Break Things" Doesn’t Work for Self-Driving Cars
Important Note: Benzinga and Yahoo Finance may earn commission from some links in the original article. This summary focuses only on the news content.
The Big Idea: Silicon Valley’s Favorite Motto Is Dangerous for Robots
Imagine you’re building a chatbot. If it gives a wrong answer, you just say "oops," fix the code, and try again. No big deal.
Now imagine you’re building a self-driving car. If it makes a mistake, people can get hurt—or worse. There’s no "undo" button in real life.
That’s exactly what Dmitri Dolgov, co-CEO of Waymo (Alphabet’s self-driving car company), told developers at a Y Combinator interview. His message was loud and clear:
"Move fast and ship safely."
Why "Move Fast and Break Things" Fails in Physical AI
The Old Way (Digital Products)
- Mistake: Chatbot hallucinates an answer
- Fix: Push an update, user tries again
- Cost: Wasted time, maybe frustration
The New Reality (Physical AI / Robots / Cars)
- Mistake: Car drives into a flooded road or construction zone
- Fix: Recall 3,800+ vehicles, suspend service, investigate
- Cost: Human lives, not tokens
Key Quote from Dolgov:
"There’s simply not an undo and a retry button."
Safety Must Be Built In—Not Bolted On
Dolgov emphasized that safety isn’t a feature you add later. It has to be baked into everything from day one:
| What Needs Safety Built In | Why It Matters |
|---|---|
| AI Models | The "brain" must prioritize safety over speed |
| Training Methods | Cars must learn from rare, dangerous scenarios before they happen |
| System Architecture | Redundancies, fail-safes, and human oversight built into the foundation |
Waymo’s Track Record: Strong, But Not Perfect
The Good News (By the Numbers)
- 220+ million rider-only miles driven
- 500,000+ fully autonomous trips every week
- Independent research (56.7M miles studied):
→ Statistically fewer injury crashes and airbag deployments than human drivers
The Wake-Up Calls (Recalls & Scrutiny)
| Incident | Vehicles Affected | What Happened |
|---|---|---|
| Flooded Roads | ~3,800 | Software allowed cars into dangerous flood zones |
| Construction Zones | ~3,900 | Cars entered closed freeway work areas (12+ incidents) |
| School Buses | Under investigation | Vehicles passed stopped school buses illegally |
| Emergency Scenes | NHTSA warning | "Clear pattern" of blocking ambulances/firefighters, ignoring cones/smoke/lights |
NHTSA Administrator Jonathan Morrison:
"Any automated vehicle unable to interact safely with first responders presents a danger to the public."
What This Means for the Future of AI
Dolgov’s warning cuts both ways:
- Waymo’s data proves cautious, safety-first deployment works (better-than-human stats)
- Waymo’s recalls prove even the best systems need constant testing, learning, and fixing after launch
Bottom Line: Physical AI isn’t software. It shares our streets, our sidewalks, and our risks. The "move fast" culture that built Facebook and Uber cannot be the culture that builds robots that drive our kids to school.
Summary
- Dmitri Dolgov (Waymo co-CEO) says Silicon Valley’s "move fast and break things" is unacceptable for autonomous vehicles and robots
- Proposed new mantra: "Move fast and ship safely"
- Safety must be foundational—in models, training, and architecture—not an afterthought
- Waymo leads the industry with 220M+ safe miles and 500K+ weekly trips, outperforming human drivers in studies
- But recalls and federal scrutiny show the tech still has dangerous blind spots (floods, construction, emergency scenes)
- Physical AI demands a higher standard because mistakes cost lives, not tokens
FAQ
What is "Physical AI"?
Physical AI refers to artificial intelligence that operates in the real world through machines—like self-driving cars, delivery robots, or humanoid robots. Unlike chatbots or image generators, their actions have physical consequences.
Why can’t we just update self-driving cars like we update apps?
We do update them over the air. But updates can’t undo a crash that already happened. In software, a bug annoys a user. In a car, a bug can kill someone. The stakes are fundamentally different.
Is Waymo safer than human drivers?
According to independent research on 56.7 million miles: Yes. Waymo’s robotaxis had statistically lower rates of injury-reported crashes and airbag deployments than human drivers in the same conditions.
What were the Waymo recalls about?
- May 2024: ~3,800 cars recalled—software could route vehicles into flooded roads
- Later recall: ~3,900 cars—entered closed construction zones (12+ incidents), leading to a temporary freeway driving suspension
What is NHTSA and why does it matter?
NHTSA = National Highway Traffic Safety Administration (U.S. auto safety regulator). They investigate defects, mandate recalls, and set safety standards. Their July warning about a "clear pattern" of autonomous vehicles interfering with emergency responders is a serious signal to the entire industry.
© 2026 Benzinga.com. This summary is for informational purposes only and does not constitute investment advice.
