OpenAI unveils progress in automated AI research

OpenAI has announced significant advancements in automated AI research, with the development of an 'automated research intern' that can perform well-defined tasks under human supervision. This achievement, announced last fall, was reached by September 2026, with further progress aimed at achieving full research automation by March 2028. The company reports that coding agents are now integral to daily research workflows, with the median researcher using over $600 per day of inference by mid-August. The research organization now uses 3.1 agent-workdays for every human workday, indicating a substantial shift in productivity. OpenAI emphasizes the importance of transparency and safety in developing these systems, citing the recent Hugging Face incident as a catalyst for increased safety measures. They have paused reinforcement learning training on their latest models and enhanced monitoring systems. The company also highlights the growing role of coding agents in troubleshooting and infrastructure improvements, with a notable decline in human-led technical support sessions. These changes reflect a broader trend toward automation in AI research, with OpenAI committed to balancing innovation with responsible development.