On September 6, OpenAI announced that it had achieved the goal set last autumn to realize an "autonomous research intern" by September 2026. An "research intern" refers to a system capable of executing clearly defined research tasks—which would typically take a skilled researcher several days—under human direction. OpenAI is now aiming to realize an "autonomous AI researcher" by March 2028.
The use of AI agents is progressing rapidly within the company's research division. As of mid-August, it is reported that the research division is utilizing agent labor equivalent to 3.1 humans per day. The number of experiments per researcher also recorded its highest level since January 2025 in August 2026.
At the same time, the company mentioned safety concerns accompanying the acceleration of research driven by AI. OpenAI acknowledged the possibility that safe control methods may not keep pace with the improvement of capabilities in "Recursive Self-Improvement (RSI)," where AI advances research autonomously. Furthermore, the company revealed an instance where it restricted the allocation of computing resources to specific models for safety reasons, following an incident where agents intruded into the research infrastructure.
Source: