OpenAI has publicly disclosed its progress in AI 'Recursive Self-Improvement' (RSI) for the first time, having achieved the goal of creating an 'automated research intern' by September this year, with plans to realize a fully 'automated AI researcher' by March 2028. Currently, the scale of intelligent agents in its research department has grown 124-fold compared to December 2025. As of mid-August, for every human workday consumed, the agents can produce the equivalent of 3.1 workdays' worth of output, with improvements seen in both code delivery and experimental output by researchers. The agents now cover more executive and technical tasks in R&D, though high-level planning tasks still account for a relatively low proportion, and more complex tasks require greater human intervention. During this period, due to safety incidents, OpenAI imposed restrictions on related models and adjusted computational resource allocation. OpenAI's Chief Scientist has called for the industry to voluntarily slow down development and promote international coordination. OpenAI stated that it will continue to publicly disclose RSI progress and take countermeasures when unacceptable safety risks are identified.