What happened
OpenAI is hiring a safety researcher for its Preparedness team, offering a pay package between $295,000 and $445,000. This role focuses on mitigating risks from "recursive self-improvement," where AI systems could enhance their own capabilities. Responsibilities include defending models from data poisoning, interpreting model reasoning, and tracking the automation of technical staff, as stated in the job listing. This move follows rapid advancements in AI, with researchers at METR noting that the length of tasks frontier models can complete autonomously doubles approximately every seven months.
Why it matters
The recruitment signals OpenAI's prioritisation of future-state risks associated with increasingly autonomous AI, shifting focus towards proactive defence against emergent capabilities. Security architects and research teams must now contend with the implications of AI systems that could independently advance, including potential vulnerabilities from data poisoning and the need for enhanced model interpretability. This follows Anthropic co-founder Jack Clark's prediction of a 60% chance that AI research and development could occur without human involvement by the end of 2028, underscoring the urgency for robust safety frameworks.
