Sam Altman has posted a job that reads less like a standard senior hire and more like a short, high-stakes experiment in corporate anxiety: OpenAI is recruiting a Head of Preparedness, a senior role tasked with imagining how modern AI systems could go wrong — and then building the tests, policies and stopgaps to prevent those outcomes. The announcement came from Altman on X and the position is listed on OpenAI’s careers page.

On paper, the head of this new function will author and run OpenAI’s internal “preparedness framework” — the playbook the company says it will use to track frontier capabilities and the novel risks those capabilities might introduce. That means owning capability evaluations, threat models and mitigation design, and turning the results of those exercises into operational safeguards and launch gates before powerful features ship. OpenAI’s posting frames it as an end-to-end responsibility: not just proposing checks, but building and enforcing them across research and product teams.

In practice, the job will be technical and managerial at once. The person in this role is expected to design tests that stress the limits of models in concrete domains — for example, whether a model can meaningfully assist in cyber-attacks, help design biological agents, or engineer large-scale manipulation campaigns — and then translate those results into hard rules for release, from gating criteria to product policy and technical mitigations. The listing makes clear this is more than an advisory ethics role: it’s meant to be a bottleneck that can say “not yet” when evaluations show unacceptable risk.

Altman himself framed the hire as a response to how quickly models have improved and how messy their side effects already look. In his post, he pointed to real-world concerns OpenAI has observed — notably mental-health harms tied to conversational agents, models that can write or debug code well enough to be useful to both defenders and attackers, and the specter of systems that autonomously improve or enable biological capabilities. The tone of the announcement is blunt: these are problems the company wants someone senior to stare at every day.

OpenAI did not hide the stakes or the incentives. News reports and the posting itself list compensation in the roughly half-million range plus equity, and Altman calls the job “stressful,” saying the successful candidate will be thrown into the deep end immediately. The role’s remit reads like a cross between a chief risk officer for frontier systems, a red-team lead and a product gatekeeper who must interpret technical evaluations and directly influence launch and policy decisions.

Taken publicly, the hire is a signal as much as a staffing decision. By codifying preparedness as a named, funded function with senior authority, OpenAI is acknowledging that traditional QA, red-teaming and content filters are not sufficient when systems introduce qualitatively new harms. It is also an attempt to show regulators, partners and the public that safety is an internal structure with career-risk attached — not merely a PR line. For competitors and regulators, the move raises the bar: if OpenAI needs a Head of Preparedness, the implication is that generic ethics teams won’t cut it for the next wave of capabilities.

The job is, in short, a formalization of a habit of structured worry. Its daily questions are simple but consequential: what can this model really do, who could exploit that ability, and what must be true before we let it loose? Whether the hire reduces real-world harms will depend on how much power the office is granted inside OpenAI, how rigorously its tests are designed, and whether the company is willing to delay or narrow launches when the answers are troubling. For now, the posting is a concrete admission that one major AI developer wants someone paid, empowered and accountable to imagine the worst and stop it from happening.