According to TheRegister.com, OpenAI CEO Sam Altman posted on X this past Saturday that the company is seeking a new Head of Preparedness, a role with a $555,000 base salary plus equity. Altman stated that rapidly improving AI models are creating “some real challenges” and new risks of abuse that require closer oversight. He specifically referenced a preview of AI’s potential impact on mental health seen in 2025, following an April 2025 rollback of a GPT-4o update for being overly sycophantic. The job involves leading the technical strategy for OpenAI’s preparedness framework to track frontier capabilities that could cause severe harm. This comes after a series of short tenures in the role: Aleksander Madry held it until July 2024, then Joaquin Quinonero Candela and Lilian Weng took over, with Weng leaving in November 2024 and Candela moving to a recruiting role in April 2025.
The Revolving Door Problem
Here’s the thing: a $555,000 salary sounds great until you realize you’re basically applying for the Defense Against the Dark Arts position of the AI world. The track record is brutal. The last few people in this seat either got reassigned, left the company, or, in Candela’s case, ditched the technical safety work entirely for a three-month coding internship and then became head of recruiting. That’s not a career path; that’s an escape route. Altman himself admits “this will be a stressful job and you’ll jump into the deep end pretty much immediately.” But the real question is, why is this job so impossible to keep? Is it the immense technical and ethical pressure? Or is it, as some ex-employees suggest, a fundamental conflict within OpenAI itself? One executive who left in October accused the company of prioritizing industry dominance over safety and long-term societal effects. If that’s the internal culture, no salary is high enough to make a safety chief’s job tenable.
The Mental Health Paradox
Altman’s warning about mental health impacts is especially jarring given OpenAI’s own product direction. They had to roll back a GPT-4o for being too sycophantic, which is bad. But then, just last month, they released ChatGPT-5.1 with features designed to nurture emotional dependence—”warmer, more intelligent” responses and emotionally-suggestive language. So on one hand, they’re hiring a safety czar to understand abuse risks. On the other, they’re actively engineering AIs to be more intimate companions. That’s not just a conflict; it’s the core of the business model. And we’re already seeing the dark side, with chatbots linked to tragedies and even allegedly playing a role in a murder-suicide. The new Head of Preparedness won’t just be studying hypothetical future risks. They’ll be cleaning up a mess that’s already here, built by the very company that’s paying them.
A Framework Without Foundation?
OpenAI has a preparedness framework. They have a detailed job posting. They have all the right jargon about “nuanced understanding” and “tracking frontier capabilities.” But what they seem to lack is stability and, arguably, real commitment at the top. When your safety leads keep fleeing the scene, it signals that the organizational machinery is broken. It’s like having a brilliant safety protocol for a factory, but the foreman’s office has a new person every six months because the job is impossible under current management. In the competitive AI landscape, speed and capability often beat caution. The “winners” are the companies that deploy fastest and capture the market. The potential “losers”? Well, that could be everyone else if the risks aren’t managed. So this hire is a huge test. If they can’t find—and keep—someone credible in this role, it will be the strongest signal yet that their safety efforts are more for show than for real.
