OpenAI reports AI "research interns" and warns about its own pace at the same time

Details

OpenAI published two documents on September 7, three days after unveiling GPT-6 Astra: an internal metrics blog detailing AI-driven research automation, and chief scientist Jakub Pachocki's essay 'An Alien Mind.' Together they represent the most detailed public account of how far recursive self-improvement has progressed inside the company. The numbers are stark. As of mid-August 2026, OpenAI's research organization logs 3.1 agent workdays for every human workday. The median researcher burns more than $600 per day in inference costs at API prices; the 90th percentile exceeds $7,000. Token output per median researcher has risen 124-fold since December 2025. Agent runtime has exceeded human working hours since June 2026. OpenAI says it has met its 'automated research intern' goal — a system that handles clearly-scoped multi-day tasks under human supervision. The next target: a fully automated AI researcher by March 2028. Pachocki's essay breaks from the usual optimistic lab framing. He states directly that no laboratory — implicitly including OpenAI — has alignment or monitoring capabilities adequate to support scaling at maximum speed. OpenAI acknowledges all figures are self-reported with no independent review, and notes that the least automatable tasks become bottlenecks, so aggregate research progress likely grows slower than individual productivity metrics suggest. The combination — record automation metrics alongside a candid warning from the chief scientist — is unusual in AI lab communications and suggests genuine internal debate about the pace of development.