OpenAI building an automated researcher
OpenAI says it's investing heavily to build a 'fully automated researcher'—an embodied agentic system that can run experiments and learn in the physical world, not just process text. The move signals a push to merge foundation models with real-world experimentation and robotics research. (technologyreview.com)
Chief scientist Jakub Pachocki spoke at length to MIT Technology Review about the new focus and framed the effort as a concentrated “North Star” program in an interview published March 20, 2026. (technologyreview.com) OpenAI has set internal milestones that aim for an “automated AI research intern” by September 2026 and a “true automated AI researcher” by March 2028, timelines CEO Sam Altman summarized on X and in company briefings. (techcrunch.com) Altman and the company say the intern target is intended to run at very large scale — described as operating on “hundreds of thousands” of GPUs — a claim tied to OpenAI’s broader compute expansion plans. (mitsloanme.com) OpenAI has been beefing up both compute partnerships and hardware supply lines, including reported multi-year GPU provisioning from AMD (capacity described as ~6GW of power draw) and large cloud deals reported with AWS to source “hundreds of thousands” of accelerators. (datacenterdynamics.com) The company is also expanding into embodied work: recent reporting and job listings document a growing OpenAI robotics team and the “OpenAI for Science” initiative that the company says will integrate models with lab and experimental tools. (venturebeat.com) Independent demonstrations of agentic research—like Andrej Karpathy’s autoresearch agents that ran roughly 700 experiments over ~48 hours and flagged model design oversights—are being cited as technical precedents for the kind of automated experiment loop OpenAI plans to scale. (msn.com) Company leaders have qualified the project’s scope, saying the system is “not to be confused with a human who researches AI” and warning internally that they “may totally fail” to meet the timetable even as they commit major resources to development. (mitsloanme.com)
