OpenAI says it hit its 'Automated AI Research Intern' milestone, targets autonomous researcher by 2028
xiaohu · x · 2026-09-07
In a new report, Research acceleration: The view inside OpenAI, OpenAI says it has met the internal milestone Sam Altman set last fall — well ahead of the September 2026 deadline — by building an "Automated AI Research Intern" that can independently complete multi-day research tasks under human supervision.
- Strict definition: not a marketing term — a system that executes well-defined, complex research tasks end-to-end under a human scientist's goal-setting and guidance, at the level of an experienced researcher's days-long work.
- Next milestone: an "Automated AI Researcher" that autonomously proposes hypotheses, designs experiments, and iterates in a closed loop, targeted before March 2028.
- Internal impact: automated research now underpins roughly 3.14x the lab's working hours, with models building eval sets, patching training clusters, and feeding back into next-generation training.
- On RSI: the report calls recursive self-improvement a physical reality already happening in the office, but takes an anti-racing stance — pursuing RSI must depend on maintaining absolute human control and democratically informed decisions about risks and benefits.
More from AGI Musings
- Predicting the industrial revolution would've looked eschatological — and it would've been right — AndyMasley · 2026-09-07
- Viral take: you don't hate AI, you hate that idea guys no longer need your permission — HankYeomans · 2026-09-07
- Andrew Critch: Voluntary Unilateral AI Slowdowns Are Underrated, Not Anti-Profit — AndrewCritchPhD · 2026-09-07
- Burkov: Software engineering will evolve, not die — 90% of coders will stop hand-writing code — burkov · 2026-09-07
- Reward hacking stems from bad reward modeling and eval awareness, developer argues — secemp9 · 2026-09-07
- NeurIPS sells out before author notification as agents snipe registration in minutes — andrewgwils · 2026-09-07