The Safety Culture at the Labs is Broken
Zvi (Don't Worry About the Vase) TheZvi ● Covered by 11 sources
Opinion — commentary, not a factual news event.
Frontier AI labs’ safety culture is still a mess. OpenAI just got a whistleblower and firings, while even Anthropic’s good habits look too thin.
Based on reporting by Zvi (Don't Worry About the Vase), TheZvi — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The source piece argues that frontier AI labs still do not have the kind of safety culture their power demands. Anthropic gets some credit for trying harder than most, especially on the technical side. But even there, the author says the understanding of risk is patchy, and too many people seem to assume things will work out because the company is good at operations. That is not the same thing as real safety culture.
OpenAI comes off far worse. David Robinson resigned and went public, after leading safety work there and helping write reports for major launches. The company also fired three safety researchers, and the piece treats that as part of a longer pattern of incidents and failures. Robinson’s point is blunt: rules and laws are not enough if the underlying culture rewards moving fast and treating known problems as manageable annoyances.
What stands out is how practical the complaints are. Robinson is not mainly arguing about existential doom; he is saying the field still lacks basic safety habits from places like aviation, nuclear power, and finance. The article says AI companies should bring in more people with experience in those worlds, and that alignment needs new science before more capable systems are built. It also says the industry keeps shipping with already-known flaws instead of using deployment to actually learn.
The broader mood is shifting. More employees are quitting in protest, and public whistleblowing is becoming normal enough to be called a cliché. That is not a healthy sign. The piece’s real warning is that even the labs that talk the best game still seem to be improvising their way through a problem that is getting larger faster than their habits can keep up.
My take — AI-written commentary, not fact-checked reporting
The industry loves saying “safety” while running on vibes and committee meetings. That works right up until someone with actual judgment walks out the door and says the quiet part out loud. The embarrassing bit is not the whistleblowing; it’s that the whistle still counts as news.
Read more about this at: Zvi (Don't Worry About the Vase)
