OpenAI reports AI "research interns" and warns about its own pace at the same time
decoder ·
OpenAI released internal data and an essay by chief scientist Jakub Pachocki, saying AI agents now play a major role in its research and that it has reached its goal of an automated 'research intern.' The company reports that the median researcher spends over $600 per day on inference, token output has jumped 124-fold since December 2025, and by mid-August agent runtime equaled 3.1 workdays per human workday. While agents handle many tasks, the report notes that tasks requiring four to eight human hours often need human intervention, and OpenAI acknowledges its metrics are hard to interpret. Pachocki warns that AI systems are becoming harder to control, that chain-of-thought monitoring is losing reliability, and that no lab has solved alignment and monitoring enough to keep scaling at maximum speed. He calls for binding standards and independent oversight, yet argues OpenAI must keep racing ahead to build defensive systems, creating a contradiction between its warnings and its pace.
decoder