OpenAI Offers $295K-$445K for Safety Researchers to Tackle AI Self-Improvement Risks

According to reports, OpenAI has posted a job listing this month on its "Preparedness safety team" for a safety researcher focused on the risks of "recursive self-improvement." The annual salary range is $295,000 to $445,000, and the company is looking for a "strong technical executor" to support its efforts in preparing for AI systems that may be able to self-improve. OpenAI has previously made clear its goal of building AI tools capable of autonomously researching ways to improve themselves.

"Recursive self-improvement" refers to an AI system's ability to autonomously train more capable versions of itself. With coding tools from OpenAI and Anthropic making significant strides over the past six months, this potential risk has drawn serious attention from industry leaders. Research institute METR published a report in March of this year showing that the duration of tasks frontier AI models can complete roughly doubles every seven months, meaning these models are gradually taking on work that would take humans far longer. This hire signals that OpenAI is actively preparing for an era of rapid AI self-evolution.

Notably, OpenAI also recently opened applications for its "Opening Eyes" residency program for the class of 2026, specifically targeting self-taught developers. The six-month full-time role offers a salary of over $220,000. The company explicitly stated that it values "learning trajectory over degrees," emphasizing that practical building skills and fast learning speed often speak louder than traditional resumes.

Looking at these moves, OpenAI appears to be pursuing a dual strategy: on one hand, hiring top-tier safety experts to guard against the risk of runaway technical development; on the other, attracting diverse talent to accelerate research. The tipping point for recursive self-improvement may be closer than many outside observers expect, and OpenAI clearly wants to stake its position ahead of time.