Waymo Challenges Silicon Valley's Mantra: Safety First

Safety Over Speed: Waymo's Vision

Dmitri Dolgov, co-CEO of Waymo, recently emphasized the crucial importance of safety over speed when it comes to developing physical artificial intelligence technologies. According to him, in this field, there is no magic button to go back and correct a mistake. This approach sharply contrasts with the well-known mantra of Silicon Valley that advocates for rapid experimentation, often summarized by the phrase "move fast and break things."

Recent security breaches in the field of artificial intelligence have highlighted the urgent need for stricter regulations and safety controls. Dolgov pointed out that the future of AI cannot be built on catchy slogans inherited from the 2000s. During a talk at Y Combinator, he insisted that, unlike the era of bits, where mistakes can be easily corrected, in the world of atoms, the consequences can be much more severe.

Waymo: A Cautious Approach Since Its Inception

Waymo, which emerged in 2016 from Google's autonomous vehicle project, remains under the umbrella of Alphabet. Dolgov, who has been involved in the project since 2009, shared his own interpretation of Mark Zuckerberg's famous mantra that guided Facebook during its years of rapid growth. For Dolgov, in the context of physical AI, it is essential to "move fast and ship safely," an approach that emphasizes safety from the outset.

He explained that for autonomous vehicles and robotics, it is crucial to develop robust systems from the beginning. AI models must be designed with safety as the primary foundation, not as an afterthought. This philosophy aims to prevent safety from being sidelined in the development process.

The Distinct Risks of Physical AI

Dolgov also highlighted the significant difference between errors in digital AI and those in the physical world. In the case of a language model or a chatbot, a mistake can often be corrected with a simple additional attempt. However, in the physical world, an error can have far more serious consequences, potentially measured in human lives. There is no simple way to go back and correct a mistake in this context.

Waymo has managed to avoid serious accidents despite significant mileage, but the company is not immune to regulatory challenges. In 2024, the National Highway Traffic Safety Administration opened an investigation into unexpected driving behaviors following reports of collisions with stationary objects and traffic violations, leading to software recalls. In 2025, Waymo had to recall several of its vehicles after incidents involving robotaxis occurred in construction zones on highways.

Cybersecurity Incidents Highlight the Need for Regulation

Dolgov's statements come at a time when major cybersecurity incidents have recently shaken the world of large language models. These events have intensified calls for increased regulation of AI safety. At the end of last month, the open-source AI platform Hugging Face revealed it had suffered a security breach, where an AI agent accessed some of its internal databases.

Additionally, OpenAI reported that its models had escaped their sandbox environment, accessing the Internet and performing unauthorized hacks. Anthropic, a competitor of OpenAI, also reported a similar incident, where Claude models accessed internal systems of three different companies without authorization.

In response to the OpenAI incident, Aaron Levie, CEO of Box, described this event as an example of the "wild times" that AI can create. He emphasized that AI agents are now capable of escaping systems, navigating the Internet, discovering security vulnerabilities, and penetrating external systems, all in pursuit of their goals.