OpenAI Offers $445,000 to Study Self-Improving AI

OpenAI is hiring a researcher for its safety division with a compensation package reaching as high as $445,000, as the company intensifies work on risks tied to AI models that may eventually improve themselves without human intervention.

According to a report by Business Insider, the role sits within OpenAI’s Preparedness team and focuses on “recursive self-improvement,” a concept where AI models could autonomously enhance their own capabilities over time.

The job listing reportedly offers a salary range between $295,000 and $445,000 and seeks candidates capable of reasoning about future threats that may not yet be fully visible. OpenAI is specifically looking for researchers who can think in a “tasteful and strategic” manner while addressing emerging AI safety risks.

Business Insider reported that the researcher would work on areas such as defending AI systems against data poisoning attacks, improving tools that interpret AI reasoning processes, and evaluating how AI could automate highly technical forms of labour.

OpenAI CEO Sam Altman has previously spoken about the company’s ambition to develop an “automated AI researcher” by 2028, while also acknowledging the technical and safety challenges involved.

The Preparedness team reportedly also examines broader risks associated with frontier AI models, including cybersecurity threats and dangerous model behaviour. The move comes amid ongoing debates around whether leading AI labs are scaling capabilities faster than safety protections.

Concerns around self-improving AI are also becoming more visible across the broader industry. 

Google DeepMind CEO Demis Hassabis recently said humanity may already be in the “foothills of the singularity,” referring to a future where AI systems improve themselves faster than humans can effectively monitor. 

Meanwhile, Anthropic has repeatedly discussed the importance of alignment and constitutional AI techniques as models become more capable.