Overview
- OpenAI targets an autonomous "AI research intern" by September 2026 as a proof of concept for tackling limited research problems.
- The company aims to scale that prototype into a multi-agent system running in data centers by 2028 to take on larger, multi‑domain projects in science, software, business, and policy.
- The envisioned agents would plan tasks, analyze information, test hypotheses, and work for hours or days with minimal human direction.
- OpenAI cites internal tools that already finish tasks once requiring days of manual work as evidence of progress toward longer-running agents.
- Safety work includes experiments to monitor model reasoning during operation, with highly capable systems potentially confined to restricted environments.