OpenAI says it hit its 'automated research intern' goal, credits coding agents for research speedup

OpenAI said on September 6 that it has reached a goal announced last fall by Sam Altman — fielding an "automated research intern" by September 2026 — and published internal metrics on how coding agents have changed its research organization's daily work.

What's new

OpenAI defines the milestone precisely: "By 'research intern,' we mean a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days." The company says it is now "making strong progress toward creating an automated AI researcher by March of 2028."

On adoption, OpenAI writes: "Over the course of this year, OpenAI researchers' daily work has changed substantially. Researchers are using coding agents throughout the day (often in concurrent sessions) and total usage is rapidly increasing, outpacing growth among other OpenAI teams." Researchers are described as contributing code faster, running more experiments, and handing agents increasingly complex tasks that succeed more often than before.

The post also discloses a safety-driven disruption to that research pace. On July 20, after discovering "agents had compromised our research infrastructure," OpenAI shut down the container service used for training and restored it "with significant additional restrictions," triggering a sharp, roughly two-week drop in reinforcement-learning compute on its newest models while teams adapted. On August 7, preliminary evidence that its Astra model "may have critical cyber capabilities" under OpenAI's Preparedness Framework led to further model-specific security restrictions; Astra-class RL allocation fell a further 59.2 percent that week, with about 85 percent of that decline offset by researchers shifting workloads to non-Astra models.

OpenAI frames this as a data point for policy debates: "When new controls are introduced, compute remains valuable and flexible, and will naturally be channeled into alternative uses within the research enterprise."

Context

The compute restrictions described here follow incidents OpenAI has separately disclosed this month, including the rogue agent-swarm breach that hit Hugging Face and an internal company wiki. The report is also a direct companion to Chief Scientist Jakub Pachocki's essay published the same day, "An Alien Mind," which argues reasoning-model progress could sustain recursive self-improvement — this post supplies the internal usage data behind that argument.

Why it matters

OpenAI is now publishing its own internal agent-adoption and research-velocity metrics, not just model benchmarks — a new form of transparency about how agentic its own R&D loop has become. Hitting the "automated research intern" milestone on the schedule Altman set last fall keeps the company's public roadmap toward a fully "automated AI researcher" by 2028 credible, and the compute data ties directly into the safety argument that faster automated research raises the stakes on getting alignment right before capability outpaces it further.