Tech Desk: OpenAI is making a bold move to secure its future by offering up to $445,000 for a single AI safety researcher. This high-stakes hire is tasked with contemplating a problem that doesn’t yet exist: an AI capable of recursive self-improvement.

The job posting, which targets “strong technical executors” with a “tasteful and strategic” mindset, sits within OpenAI’s Preparedness safety team. The substantial salary, comparable to senior ML engineer roles at frontier labs, underscores the gravity with which OpenAI views this position. The emphasis on being “tasteful and strategic” stems from the nature of the work – reasoning about future risks and modelling hypothetical threats, where expert judgment is as crucial as coding prowess.

This researcher will work on critical tasks such as defending models from data poisoning, developing interpretability tools, and tracking AI’s progress in automating technical roles within OpenAI itself. The broader Preparedness team also tackles automated red-teaming, bio/chemical risks, and agentic AI threats.

Recursive self-improvement, where an AI designs and trains better versions of itself, is moving from theory to reality. CEO Sam Altman aims for an “automated AI research intern” soon and a “true automated AI researcher” by early 2028. Industry leaders like Google DeepMind’s Demis Hassabis see humanity approaching the “foothills of the singularity,” with estimates suggesting AI could handle R&D without human input by 2028. The pace of AI development, with task lengths doubling every seven months, highlights the urgent need for such specialized expertise.