A job posting by OpenAI has drawn attention for both its compensation and its unusually candid language about the risks the company is preparing for. OpenAI has posted a job listing for a safety researcher focused on recursive self-improvement, the scenario in which AI systems begin meaningfully accelerating their own capabilities, with total compensation reaching $445,000 per year.
The role sits within OpenAI's Preparedness team and asks candidates to study problems that "might exist in the future, but might not exist now." The posting specifies that the work involves tracking progress toward the automation of technical staff, implementing data-poisoning defences, and designing experiments to understand the extent to which AI models may be misaligned or developing dangerous capabilities ahead of corresponding safety measures.
The listing has attracted attention beyond the AI industry for one phrase in particular. OpenAI states that because the work involves reasoning about uncertain, forward-looking risks, it is "especially important that people in this role are tasteful and strategic."
The phrase has been interpreted as a signal that OpenAI wants a researcher capable of navigating internal politics around a subject that could attract unwanted regulatory scrutiny or negative press coverage if handled carelessly.
Recursive self-improvement is the idea that AI systems can keep improving themselves after training, a growing focus at leading AI labs. Google DeepMind’s Demis Hassabis has said models may eventually continue learning after deployment, while OpenAI’s Sam Altman has spoken about building an automated AI researcher by 2028.
OpenAI is placing this work under its safety team, treating it as a potential risk to study and control rather than a capability goal. The $445,000 package is highly competitive and may push rivals like Anthropic and Google DeepMind to raise compensation for similar talent. The move also comes amid ongoing scrutiny of OpenAI’s safety structure and recent changes to its internal alignment teams and programs.
It is unclear whether the researcher will report to OpenAI’s Preparedness team or another safety group, and how their findings will be shared publicly. It also remains unknown whether the role was prompted by specific internal AI capabilities or broader precautionary planning.
If the work stays internal, outside researchers may have limited visibility into how OpenAI is assessing key AI risks.