OpenAI is hiring a safety researcher at a salary of $295,000 to $445,000 to work on one of the most consequential problems in AI development: what happens when a model can train better versions of itself. The role sits within the company’s Preparedness team, the group tasked with preventing severe harms from AI systems, according to Business Insider, which first reported on the listing.
The job posting, which appeared on aggregator sites this month, seeks “strong technical executors to support preparations for recursive self-improvement.” It includes an unusual qualifier: “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.”
What the Role Covers
The successful candidate could work across several areas, according to the listing details reported by Livemint. These include defending OpenAI’s models against data poisoning (attempts to corrupt a model through its training dataset), building tools to interpret models’ internal reasoning, and running experiments to understand the safety implications of self-improving systems. The researcher may also “track progress toward automation of technical staff,” including measuring how extensively AI coding tools are being used inside OpenAI itself.
Other open roles on the Preparedness team cover automated red-teaming for cybersecurity, biological and chemical risks, and threats from agentic AI systems. “This is urgent, fast-paced work that has far-reaching implications for the company and for society,” the postings state, per Business Insider.
The Self-Training Timeline
The posting arrives against a backdrop of explicit internal timelines. CEO Sam Altman said in October 2025 that OpenAI had set goals of running an “automated AI research intern” on hundreds of thousands of chips by September 2026, and building 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.”
Researchers at METR, a laboratory that studies AI model capabilities, wrote in March that the length of a task frontier AI models can complete doubles roughly every seven months, per Business Insider. The implication is that AI agents will soon handle a “large fraction” of software work that takes human coders days or weeks.
Google DeepMind CEO Demis Hassabis said this week that humanity now stands at the “foothills of the singularity,” the point at which AI begins to improve itself faster than human intelligence can follow.
Anthropic Is Working the Same Problem
OpenAI is not alone. Anthropic published research in April on using AI models to oversee more powerful AI models, with results described as promising but limited, according to Livemint. In May, Anthropic co-founder and policy head Jack Clark wrote that he believes there is roughly a 60% chance of AI research and development conducted without human involvement by the end of 2028.
METR CEO Elizabeth Barnes wrote on Friday that in her view, “any ‘reasonable’ civilization would clearly be taking things much more slowly and carefully with AI.”
From Research Budget to Revenue Risk
The $445,000 ceiling for a single safety researcher puts a concrete price on what has historically been a philosophical debate. When a company budgets nearly half a million dollars for someone to reason about risks “that may not exist yet,” it signals that internal research roadmaps have moved recursive self-improvement from speculative to plausible.
For agent builders and teams deploying autonomous systems, the signal is worth watching. If METR’s task-doubling timeline holds, the systems these researchers are preparing for could arrive while OpenAI is still a public company, forcing real-time decisions about capability releases against shareholder pressure. The gap between “tasteful and strategic” safety research and production deployment timelines may be narrower than the job listing implies.