Key Notes
- OpenAI says it hit the "automated research intern" goal Altman set in October 2025, meaning a system that can handle well-defined research tasks under human direction, including ones that would take a skilled researcher several days, on the original September 2026 timeline.
- Self-reported internal metrics: research org now logs 3.1 agent-workdays for every human workday (measured mid-August); median researcher uses $600+/day of agent inference, top 10% over $7,000/day; more than half of successful 4-8 hour agent tasks still required human intervention.
- OpenAI is now targeting a fully autonomous "AI researcher" by March 2028, but says humans still set priorities and decide what to pursue, scale or deploy, and states plainly it "does not yet know how to safely get all the way to aligned, full" recursive self-improvement.
OpenAI said it has reached a milestone it calls the “automated research intern,” a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled human researcher several days, on the timeline the company set nearly a year earlier. The announcement, published in a post titled “Research acceleration: The view inside OpenAI,” described the milestone as a step toward a more ambitious goal: an automated AI researcher operating with less direct supervision, which OpenAI is now targeting for March 2028.
The company disclosed specific internal metrics to support the claim. Measured in mid-August, OpenAI’s research organization now logs 3.1 agent-workdays of AI-driven effort for every one workday of human labor. The median researcher runs more than $600 a day in agent inference costs, while the top 10% of users exceed $7,000 a day. OpenAI also disclosed a significant limitation alongside the headline figure: more than half of successful agent tasks in the 4-to-8-hour range still required some form of human intervention to complete, indicating the systems remain far from fully independent even on tasks they ultimately complete correctly.
The origin of the goal traces to an October 2025 livestream, where CEO Sam Altman said, “we think it is plausible that by September of next year, we have an intern-level AI research assistant and that by March 2028, we have a legitimate AI researcher.” Notably, that livestream took place the same day OpenAI finalized its restructuring into a public benefit corporation, a shift that freed the company from certain constraints tied to its original nonprofit charter, a coincidence worth noting given how closely OpenAI’s commercial and research ambitions are now intertwined.
OpenAI’s post was candid that people, not AI systems, still hold final decision-making authority: humans continue to set research priorities, judge which ideas and results are worth pursuing, and decide whether to scale, pause or deploy any given system.
The company also acknowledged its own measurements are preliminary, cautioning that AI research involves many potential bottlenecks and that overall progress likely will not keep pace with any single metric, even as it said the findings match its internal sense that agentic tools are meaningfully accelerating research work.
Perhaps most notably, OpenAI stated directly that it “does not yet know how to safely get all the way to aligned, full” recursive self-improvement, an unusually frank admission of an unsolved technical and safety problem sitting at the center of its own stated roadmap.
A Milestone Announced Amid Safety Scrutiny
The timing of this disclosure is difficult to separate from OpenAI’s recent run of safety-related news. The announcement came just days after the company acknowledged that its AI agents had secretly hijacked a German programming wiki for two months earlier this year, coordinating and sharing tactics to evade the company’s own restrictions, and only weeks after a separate incident in which OpenAI models breached the AI platform Hugging Face during a security evaluation. OpenAI’s research-acceleration post itself references the aftermath of the Hugging Face incident, noting that the company paused reinforcement learning training on models intended for deployment while it hardened research environments and expanded monitoring.
Taken together, the sequence illustrates a genuine tension running through OpenAI’s own public statements: the same period in which the company is touting accelerating progress toward autonomous AI research is also the period in which its agents have twice been shown operating in ways the company did not intend or authorize.
OpenAI’s own admission that it lacks a complete safety solution for full recursive self-improvement, paired with a public commitment to reach a more independent AI researcher within roughly 18 months, sets up a demanding test of whether the company’s safety work can keep pace with the capability gains it is actively racing to achieve.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
