OpenAI Seeks Researcher for Self-Improving AI
Quick Look
- OpenAI is hiring a researcher for its Preparedness safety team to focus on AI that can improve itself, offering up to $445,000.
- The role requires strong technical skills and strategic judgment to model future risks and potential AI threats.
AI-generated summary
Why It Matters
OpenAI is seeking a specialized researcher to focus on the safety and potential risks of AI systems that can recursively improve themselves. This role is part of the company's Preparedness safety team and involves modeling future, unmanifested AI risks.
OpenAI is willing to pay up to $445,000 for a single researcher whose job is to think about a problem that may not exist yet: an AI that can make itself smarter. The listing, first reported by Business Insider, sits on the company's Preparedness safety team and asks for "strong technical executors" who are also "tasteful and strategic. " The pay band starts at $295,000 and tops out at $445,000. That is senior ML engineer money at frontier labs, which tells you how seriously OpenAI is treating this hire.
Why the OpenAI job ad asks for someone 'tasteful and strategic'
The phrase is odd, and the posting explains why it's there. The work involves reasoning about problems that may exist in the future but not yet today. The person will be modelling risks that have not shown up, so judgement counts as much as code. Per Business Insider, the researcher could defend models from data poisoning, build tools to interpret model reasoning, and "track progress toward automation of technical staff" — effectively measuring how much of OpenAI's own work AI is already doing. The Preparedness team's wider brief covers automated red-teaming, biological and chemical risks, and agentic AI threats.
Recursive self-improvement is no longer just a thought experiment
The idea is simple and unsettling: an AI that researches, designs and trains better versions of itself, needing less human input each cycle. CEO Sam Altman has been blunt about the company's goals. He wants an "automated AI research intern" running on hundreds of thousands of chips by September, and a "true automated AI researcher" by March 2028. He is not alone. Google DeepMind chief Demis Hassabis said this week that humanity is at the "foothills of the singularity." Anthropic policy head Jack Clark puts the odds of AI doing R&D without humans by end-2028 at roughly 60 per cent. METR researchers found the task length frontier models can handle doubles every seven months.
What to Watch
AI outlook — possibilities, not facts
OpenAI will likely fill the specialized researcher position due to the high compensation offered.
Very likely · Within months
The focus on AI self-improvement will intensify competition among leading AI labs.
Likely · Short term
Open Questions
- What specific methodologies will the researcher employ to model non-existent AI risks?
- How will OpenAI measure progress in AI self-improvement and automation?
- What are the potential ethical implications of developing self-improving AI?
- What is the timeline for the 'automated AI research intern' and 'true automated AI researcher' goals?
