OpenAI to Allocate ₹3.7 Crore for Research on Preventing Autonomous AI Development

OpenAI Seeks Researcher for Critical AI Challenge: Recursive Self-Improvement

OpenAI, a leading artificial intelligence company, has posted a job listing for a researcher to work on one of the most pressing challenges in the field: recursive self-improvement. This concept, where an AI system can research, design, and train better versions of itself without human involvement, has become a top priority for the industry over the past six months.

According to the job listing, the researcher will be part of OpenAI's Preparedness safety team and will work on several areas, including defending the company's models against data poisoning, building tools to interpret models' reasoning, and running experiments to understand the safety implications of self-improving systems. The successful candidate will also be responsible for tracking progress toward automation of technical staff, including measuring the extent to which AI coding tools are being used within the company.

The job listing comes with a salary range of $295,000 to $445,000 (approximately ₹2.5 crore to ₹3.7 crore) and requires the candidate to be "tasteful and strategic" in their work, as the role involves reasoning about problems that may exist in the future but may not exist now.

Read also: Treasury Yields Experience Largest Increase in Two Weeks Following Release of Labor Market Data

The Growing Concern of Recursive Self-Improvement

The concept of recursive self-improvement has gained significant attention in recent months, with OpenAI and Anthropic advancing their coding tools at a pace that has surprised even their own researchers. Google DeepMind chief executive Demis Hassabis has warned that humanity stands at the "foothills of the singularity," the point at which AI begins to improve itself and outpaces human intelligence.

Researchers at METR, a laboratory that studies AI model capabilities, have estimated that the length of a task that frontier AI models can complete doubles roughly every seven months. This implies that AI agents will soon be able to handle a "large fraction" of the software work that takes human coders days or weeks to complete.

OpenAI's Ambitions in Automated AI Research

Read also: US-Iran Tensions Spark Uptick in Oil Prices Amid Global Market Decline

OpenAI chief executive Sam Altman has been explicit about the company's ambitions in this area. In October, Altman said that the company had set a goal of running an "automated AI research intern" on hundreds of thousands of chips by this coming September, and a "true automated AI researcher by March of 2028."

"We may totally fail at this goal," Altman wrote on X, “but given the extraordinary potential impacts we think it is in the public interest to be transparent about this.”

Comparison of AI Model Capabilities

Note: The table shows the estimated task completion time for frontier AI models and human coders. The length of a task that frontier AI models can complete doubles roughly every seven months.

Related Open Roles at OpenAI

The Preparedness team's broader mandate includes preventing severe harms from AI. Other open roles on the same team cover automated red-teaming to test cybersecurity vulnerabilities, biological and chemical risks, and threats posed by agentic AI systems.

"This is urgent, fast-paced work that has far-reaching implications for the company and for society," the Preparedness postings say.

Anthropic's Similar Focus on Automated AI Research

OpenAI is not alone in preparing for this shift. In April, Anthropic published research on using AI models to oversee more powerful AI models, with results described as promising but limited. Anthropic co-founder and policy head Jack Clark has estimated that there is roughly a 60 per cent chance of seeing AI research and development conducted without human involvement by the end of 2028.

METR chief executive Elizabeth Barnes has expressed caution, writing that in her view, “any 'reasonable' civilization would clearly be taking things much more slowly and carefully with AI.”