OpenAI's 'Third Phase' Plan Targets an Automated AI Researcher by March 2028

OpenAI's 'third phase' plan

OpenAI has published a strategic document it calls its "third phase," whose clearest milestone is an automated AI researcher: the company says it wants AI systems to carry out a substantial share of its research by March 2028. According to dev.to, which summarized the plan, the document is dated June 8, 2026 and was written by Sam Altman and Jakub Pachocki.

The plan is not a product launch. It ships with no pricing, no rollout schedule and no new business software. Instead, as dev.to reports, OpenAI argues that raw frontier capability is not enough on its own — the harder job is converting that capability into tools people can actually use, while keeping systems safe, aligned with human intent and genuinely overseen by people.

The three commitments

The plan rests on three connected commitments, according to the dev.to breakdown:

• Build an automated AI researcher that helps speed up scientific work, with March 2028 named as the aspirational date by which AI could handle a large portion of research.
• Use AI to accelerate scientific progress and productivity, while working to ensure the resulting gains are widely shared rather than concentrated.
• Pursue "personal AGI": advanced AI that is accessible to everyone on Earth and helps individuals pursue their own goals in their own way.

The phrasing matters here. As dev.to notes, the 2028 target is a stated ambition, not a claim that autonomous research at that scale exists today or has been released to customers.

Power, oversight and open questions

The document also reportedly engages with the danger of power pooling among a small set of actors, linking broad deployment to public oversight, coordination across countries, resilience and continued safety investment as frontier systems grow more capable. That framing, according to the article, makes this primarily a strategic and mission-level statement rather than a roadmap for any particular ChatGPT feature, API or commercial bundle.

Large questions are left open. dev.to points out that the plan does not say what tasks the automated researcher would take on, how much autonomy it would have, which scientific domains it would touch, or how human researchers would supervise it. It also does not clarify whether any eventual capability would arrive via ChatGPT, an API, a standalone product or an internal system reserved for OpenAI's own researchers.

Why it matters

An automated AI researcher is a step beyond models that draft copy, summarize documents or answer questions — it points at AI being applied to the practice of research itself, with the potential to change how quickly science moves. The March 2028 date also hands the industry a public yardstick: every subsequent OpenAI release and research update can now be measured against a goal the company set for itself.

For organizations, the near-term reading should stay modest. The plan signals that OpenAI expects AI to take on more involved, multi-step knowledge work over time, but it says nothing about availability, cost or integration. Purchasing and implementation decisions should rest on what exists today, not on an assumed 2028 offering. The next meaningful signals, as dev.to observes, will be product-level: documentation covering data handling, human supervision and task boundaries, and clarity on whether the capability stays internal, ships externally or rolls out only in limited form.