OpenAI Is Offering Rs 3.7 Crore For A Job To Stop AI From Becoming Too Smart

OpenAI is offering up to $445,000 (around Rs 3.7 crore) for a specialised research role focused on one of the most debated questions in artificial intelligence: what happens if AI systems begin making themselves smarter without human involvement? The position, first reported by Business Insider, is part of OpenAI's Preparedness safety team, which studies long-term risks posed by increasingly powerful AI systems. The listing has drawn widespread attention for its unusual requirement that candidates must be not only technically strong but also "tasteful and strategic."

"This work relies on reasoning about problems that might exist in the future but might not exist now. So it's especially important that people in this role are tasteful and strategic," the listing says.

Tasks of the Role

According to the job description, the role involves reasoning about threats that may not yet exist but could emerge rapidly as AI capabilities advance. The listing centres on a concept known as "recursive self-improvement," where an AI system could potentially research, design, and train more capable versions of itself with minimal human oversight.

• The company says researchers in the role would help anticipate and mitigate risks linked to highly autonomous AI systems before they become real-world problems.
• Responsibilities include studying speculative future threats, running safety experiments on advanced models, and building systems to monitor unexpected AI behaviour.
• The role also involves defending models against data poisoning attacks, where malicious or manipulated information corrupts training datasets, as well as developing tools that help researchers understand how advanced AI systems reason and make decisions.
• One particularly notable aspect of the position is its focus on tracking how automation could begin replacing technical work inside OpenAI itself.

Recursive self-improvement

The idea of recursive self-improvement has increasingly become a serious topic within the AI industry as tools developed by companies such as Anthropic and OpenAI continue advancing at a rapid pace.

This week, Demis Hassabis, chief executive of Google DeepMind, said humanity is now standing at the "foothills of the singularity"- a theoretical point where AI systems begin improving themselves faster than humans can keep up with.

Meanwhile, researchers at METR, a lab studying frontier AI capabilities, recently claimed that the complexity of tasks AI systems can complete is doubling roughly every seven months. The group suggested that AI agents may soon be capable of handling a significant share of software engineering work that currently takes human developers days or even weeks to finish.