AgentsSeptember 7, 2026via The Decoder
OpenAI reports AI "research interns" and warns about its own pace at the same time
Why it matters
OpenAI has deployed autonomous agents doing real research work at measurable scale, hitting an internal milestone ('research intern'). But Pachocki's warning signals a frontier-lab constraint: capability is running ahead of safety/monitoring infrastructure — a critical tension for practitioners evaluating agent deployment risk.
Key signals
- OpenAI agents handle 3.1 workdays per human workday in internal research
- OpenAI claims it has achieved 'automated research intern' goal
- Chief scientist Pachocki warns alignment and monitoring infrastructure lags behind scaling capability
- Pachocki: no lab has sufficient grip to scale at 'maximum speed' safely
- Agents deployed doing real multi-step research work (not pilots)
- Published Sep 2026
The hook
3.1 workdays per human workday. OpenAI's research agents are already doing the work — but the chief scientist says nobody knows how to safely scale it.
OpenAI reports that AI agents in its own research already handle 3.1 workdays for every human workday, and it says it has reached its goal of an "automated research intern." But chief scientist Pachocki warns that no lab has a good enough grip on alignment and monitoring to keep scaling at maximum s…