OpenAI offers up to $445,000 role

OpenAI is advertising a research job that offers a listed base salary of $295,000 to $445,000 for work on one of the company’s most explicit safety concerns: AI systems that could improve themselves. The live posting, titled “Researcher, Recursive Self-Improvement Preparedness,” appears in OpenAI’s hiring system and says the role sits within its Preparedness work. (jobs.ashbyhq.com) The listing has drawn attention because it spells out a technical risk that AI companies have usually discussed in broader terms. It also shows OpenAI continuing to staff up a unit dedicated to extreme-risk scenarios. ### What is OpenAI actually hiring for? The job posting names the role “Researcher, Recursive Self-Improvement Preparedness” and lists an estimated base salary range of $295,000 to $445,000. (openai.com) OpenAI says total compensation also includes equity, performance-related bonuses for eligible employees and benefits. The application asks whether candidates can work from a U.S. office three days a week. The role’s title matters because “recursive self-improvement” is the term OpenAI is using for a specific category of frontier-model risk. In plain terms, that means studying systems that might help create stronger versions of themselves or otherwise accelerate their own capabilities. (jobs.ashbyhq.com) That framing comes from the company’s own listing, not outside commentary. ### Why does the phrase “recursive self-improvement” stand out? OpenAI’s Preparedness materials say the company tracks advanced AI capabilities that could create “risks of severe harm.” In its updated Preparedness Framework, OpenAI says it is focusing on safeguards for future models that are more capable than current ones and defines severe harm as outcomes such as thousands of deaths or hundreds of billions of dollars in economic damage. (jobs.ashbyhq.com) The company’s broader Preparedness work dates back at least to October 2023, when OpenAI said it was building a Preparedness team to support the safety of highly capable AI systems. That earlier statement said the effort was aimed at catastrophic-risk preparedness, giving context for why a role tied to self-improving systems would sit inside the same organization. (jobs.ashbyhq.com) ### Where does this role sit inside OpenAI’s safety structure? OpenAI’s careers pages show Preparedness as a dedicated hiring category with multiple open roles. Other current listings include “Security Researcher, Agentic AI Threats,” “Researcher, Frontier Biological and Chemical Risks,” “Researcher, Frontier Cybersecurity Risks,” and “Researcher, Automated Red Teaming.” (cdn.openai.com) The “Agentic AI Threats” posting describes Preparedness as “a critical Safety Research team” focused on mitigating AI threats to global security that could scale to an extreme level of severity. It says the team’s work includes measurement, mitigation and coordination, including maintaining OpenAI’s preparedness framework. (openai.com) ### What does the pay range tell us? The compensation figure in the posting is a base-salary range, not a guaranteed total package. OpenAI says in the listing that actual pay can vary by market location, job-related knowledge, skills and experience, and that equity and bonuses may be added on top. (openai.com) The number is still notable because it places safety work in the same compensation band as other highly specialized frontier-AI roles. OpenAI’s current Preparedness listings show similarly high ranges for adjacent work, including $293,000 to $405,000 for the agentic-threats security researcher role. (openai.com) ### Where can readers verify the details? The live Ashby-hosted posting contains the salary range, benefits language and application details. OpenAI’s careers search page shows the role within a wider Preparedness hiring push, and OpenAI’s published Preparedness Framework explains how the company describes frontier risks and the safeguards it says it is building around them. (jobs.ashbyhq.com) As of May 24, 2026, those pages remained publicly accessible on OpenAI-operated or OpenAI-linked sites. (openai.com)