At OpenAI DevDay on September 29, 2026, Sam Altman said the company had reached its September goal for an AI research intern a few weeks earlier. OpenAI had described the goal as met in a September 6 account of its internal measurements. The company defines the intern as a system that performs well-defined research tasks under human direction; its reported figures track agent runtime and human involvement, not the quality of research discoveries.
What OpenAI means by an AI research intern
OpenAI uses the term for a system that can carry out well-defined research tasks under human direction, including tasks that could take an experienced researcher several days. People still set research priorities, assess ideas and results, and decide whether to expand, pause, or deploy systems.
That definition describes supervised task work. It does not mean the system sets its own research agenda.
Agent runtime and human intervention inside OpenAI
By mid-August 2026, OpenAI reported 3.1 agent workdays of runtime for every human workday in its research organization. The company used an eight-hour day as the reference. This is a measure of how long agents ran, not of completed research output or scientific quality.
OpenAI also reported that more than half of its successful tasks estimated to take 4–8 hours required at least one human intervention during the preceding six months. The runtime ratio and intervention finding measure different things: one tracks agent runtime, while the other describes human involvement in a particular group of tasks.
OpenAI’s next stated research goal
OpenAI’s next stated target is an automated AI researcher by March 2028.
