OpenAI says it has reached its “automated research intern” milestone

What OpenAI reported

OpenAI says it has met an internal goal announced in 2025: an “automated research intern” able to carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days. The company is not claiming that the system independently performs the full role of a researcher. OpenAI says people still set priorities, judge which ideas and results to pursue, and decide whether to scale, pause or deploy systems.

The company also published internal usage measures showing how quickly agent-assisted work has grown. It says that, by mid-August, its research organisation was using 3.1 agent-workdays for every human workday, measured against an eight-hour day. OpenAI reports that its median researcher was using more than US$600 per day of agent inference at API prices, while the 90th-percentile user exceeded US$7,000 per day. These figures are company measurements, have not been independently audited, and do not directly establish equivalent productivity or scientific output.

The safety warning published alongside it

In a separate essay, OpenAI chief scientist Jakub Pachocki said he expects AI progress could continue into recursive self-improvement, where machine intelligence plays an increasing role in developing successor systems. He wrote that no lab has solved alignment and monitoring well enough to continue responsibly scaling at maximum speed for much longer.

Pachocki called for shared safety thresholds, third-party or government enforcement, and international coordination. He also said voluntary slowdowns should become commonplace when confidence in safeguards is insufficient. The essay is his stated assessment, not evidence that recursive self-improvement has already occurred.

OpenAI says it paused reinforcement-learning training on some models after the recent Hugging Face incident while it strengthened research environments and monitoring. It says some workloads later resumed under stronger controls while others remained paused.

What remains uncertain

OpenAI has not published the task-level evaluation set or enough independent evidence for outsiders to reproduce the “research intern” milestone. The relationship between agent runtime, research productivity and future model capability therefore remains uncertain. Independent coverage has focused on the tension between OpenAI’s reported acceleration and its chief scientist’s warning, but the central measurements currently come from OpenAI itself.

Sources

Primary source: OpenAI — Research acceleration: The view inside OpenAI · 6 September 2026

Primary source: OpenAI — An Alien Mind, by chief scientist Jakub Pachocki · 6 September 2026

Independent coverage: The Next Web · 6 September 2026

Independent coverage: Business Insider · 6 September 2026

Editorial note

SEPIDRA’s AI News & Signals section is informational. OpenAI’s measurements and forward-looking assessments are attributed to the company or its chief scientist. No external image has been reproduced; the visual above is an original CSS-based diagram.