OpenAI offers $445,000 safety job to prepare for self-improving AI

OpenAI is offering as much as $445,000 for a Preparedness team researcher focused on recursive self-improvement, putting a near-half-million-dollar price tag on the company’s push to anticipate what happens if its models start helping build the next generation of models. The opening calls for “strong technical executors” who are also “tasteful and strategic,” language that shows OpenAI wants more than raw engineering horsepower. It wants judgment, risk sense and the ability to spot problems before they become products.

The job sits inside OpenAI’s Preparedness effort, which the company announced on October 26, 2023, as a way to assess, evaluate and probe AI models for catastrophic risks. Alongside that launch, OpenAI introduced a Preparedness Challenge to pull in outside ideas and talent. The company updated its framework on April 15, 2025, saying it is tracking and preparing for advanced AI capabilities that could introduce new risks of severe harm. In that framework, severe harm means the death or grave injury of thousands of people or hundreds of billions of dollars in economic damage.

That makes the new role more than a specialist research job. OpenAI says the work is tied to preparing for recursive self-improvement, the idea that AI systems could contribute to making future systems more capable. The posting points to potential focus areas such as AI R&D risk measurement, a sign the company is thinking about how automation could move from writing code to accelerating technical research itself. OpenAI has also publicly said it is researching how to safely develop AI capable of recursive self-improvement, while its safety materials describe safety as an ongoing process of anticipating, evaluating and preventing risk.

The urgency behind that effort is easy to see in the numbers OpenAI has cited from METR, which found that frontier agents’ task length has been doubling about every seven months. If that pace continues, systems that can carry out longer and more complex software work could arrive quickly, raising the stakes for oversight, security and alignment. OpenAI’s Preparedness Framework says an internal Safety Advisory Group oversees the process and recommends safeguards before deployment, a reminder that the company is treating safety as a board-level operational concern, not a side project.

The salary itself may be the clearest signal of all. High-end compensation like this suggests the market for AI safety talent has tightened as companies race to build more capable systems while also trying to prove they can contain them. In practice, OpenAI is paying for preparedness before the technical frontier moves again, and that may be the new cost of competing at the edge of AI.