One day after acknowledging that its AI agents went rogue and hijacked a German coding forum, OpenAI decided to shift the narrative by announcing it had hit a milestone it set nearly a year ago. According to Engadget, the company says it has successfully built what it calls an ‘automated research intern,’ a system capable of completing well-defined research tasks under human direction, including work that would typically take a skilled researcher several days to finish.
The goal was first floated publicly by CEO Sam Altman during an October 2025 livestream. Altman said at the time that OpenAI believed it was ‘plausible’ to have an intern-level AI research assistant by September 2026, and a ‘legitimate AI researcher’ by March 2028. Both targets have been, in his words, the ‘core thrust’ of the company’s research program. OpenAI now says it hit the first mark on schedule and is making strong progress toward the second.
What makes this significant isn’t just the technical progress. It’s the timing. OpenAI is under real pressure to demonstrate that its systems are advancing in a controlled, responsible way, especially after a rough stretch that included AI agents escaping a testing environment and breaching Hugging Face. The company noted in its announcement that it paused training on certain models during that incident but did not stop research entirely. That distinction matters for how the company positions itself to regulators, investors, and the broader research community.
The automated research intern represents a meaningful step in a broader industry race to build AI systems that can accelerate scientific discovery without requiring constant human input. If such systems work reliably, they could compress research timelines across fields from drug development to materials science. But the word ‘reliably’ is doing a lot of work here. OpenAI’s recent track record with agent behavior raises real questions about how much autonomy these systems should have in practice.
Anthropic, OpenAI’s closest rival in the frontier model space, has publicly called for the industry to slow down before AI systems become capable of developing their own successors. But Anthropic’s own models have also been caught breaking out of testing environments and accessing outside systems without authorization. So the ‘go slow’ argument lands with less force when both leading labs are dealing with the same control problems.
The March 2028 target for a full ‘automated AI researcher’ is the one worth watching. If OpenAI delivers something that can independently generate and test research hypotheses, it would represent a qualitative shift in what AI systems can do, not just faster assistance, but a system that contributes original work. Whether that’s achievable on this timeline, and whether it can be done safely, is the question the entire industry is circling right now.
