Here's a plot twist nobody saw coming: OpenAI just dissolved the very team tasked with stopping AI from becoming a bioweapon or a cyber-superweapon. The Preparedness team, which was supposed to catch catastrophic risks before they hit the real world, is history. And the timing? Oh, just a few months before the company's massive IPO. Coincidence? I don't buy it.
Let me paint you a picture. Imagine a bank hiring a security chief, then firing the entire security department five months later—right before the bank goes public. That's basically what happened. Dylan Scandinaro, poached from rival Anthropic in February 2026 to head Preparedness, watched his unit get axed in late July. He's been handed a consolation prize: studying 'recursive self-improving AI.' Sounds important. But it's not the same as having a team that can slam the brakes on a dangerous model release.
Here's the twist: OpenAI says the work 'isn't going away.' It's being 'absorbed' into other teams. No layoffs, they promise. But let's be real—when a safety team is folded into the product org, its power evaporates. A dedicated unit whose job is to say 'no' to a model release is an obstacle to revenue. A safety person buried in a product team is just a voice that can be overruled. This isn't a reorganization; it's a dilution.
And the pattern is damning. This is the third safety-focused unit dissolved in two years. The Mission Alignment team went down in February 2026. Now Preparedness. Big-name safety leaders like Chloé Bakalar and Johannes Heidecke have already walked. The brain drain is real, and the message to the rest of the team is clear: your job isn't a priority.
The stakes couldn't be higher. The Preparedness team was responsible for assessing risks like catastrophic cyber-attacks, biosecurity threats, and rogue autonomous behavior. These aren't theoretical doomsday scenarios—they're the stuff of policy debates and front-page headlines. By dismantling this unit, OpenAI is telling the world it will handle these risks with a generalist approach. That's like replacing a fire department with a few volunteers who have buckets.
So who wins? OpenAI's IPO underwriters, who get a leaner org chart with fewer red flags for investors. Who loses? All of us. The researchers who wanted to keep us safe. And Anthropic, which lost Scandinaro to a company that just proved it doesn't value his expertise.
Here's what I'm watching next: the S-1 filing. If OpenAI's IPO documents hedge on risk management, we'll know the dissolution was a real downgrade. And Scandinaro—if he actually produces risk assessments that constrain OpenAI's roadmap, maybe this was cosmetic. If he goes quiet, the safety function is effectively dead. History says the quiet path is more likely.
The real question is whether we've learned anything from the past. We've seen what happens when companies prioritize speed over safety—in finance, in aviation, in tech. OpenAI's move could be a reformation or a burial. The fact that the people who used to work on the Preparedness team aren't the ones speaking publicly says everything. They know what's happening. And they're not optimistic.