Overview
- In May 2026 OpenAI posted a Preparedness‑team job seeking a researcher to anticipate and mitigate risks from “recursive self‑improvement,” offering a pay band of $295,000 to $445,000.
- The role lists concrete duties including defending models from data‑poisoning, building interpretability tools to read model reasoning, running safety experiments on model behavior, and measuring how much technical work AI is automating inside the company.
- OpenAI’s Preparedness team already covers automated red‑teaming for cybersecurity, biological and chemical misuse risks, and threats from agentic autonomous systems, and the new listing frames the work as urgent and strategic.
- The hire reflects broader industry concern driven by capability studies and executive statements that frontier models are improving fast and could increasingly automate research tasks, but the job is preparatory and does not indicate that fully self‑improving systems now exist.
- The move may raise practical effects for researchers and engineers as companies compete for senior safety talent and as firms plan tools and governance to manage future automation of technical roles.