On the same day it secretly filed for its IPO, OpenAI unveiled its Phase Three roadmap, establishing three key goals: developing automated AI researchers, driving economic growth, and providing every person on Earth with a personal AGI. The roadmap indicates that by March 2028, AI systems may collaborate with human researchers to complete scientific work. The company emphasized that full automation is not the goal—that as AI becomes more powerful, the role of humans becomes even more critical—and called for the establishment of international coordination mechanisms as a safety brake. The timeline for AGI remains uncertain, with industry estimates suggesting it could be achieved within the next five years.Article author and source: AI New Era
OpenAI enters Phase Three: Three Key Goals Released in Sync with IPO
Recently, OpenAI officially entered its third phase.
OpenAI CEO Altman posted OpenAI's latest plans on X.
The plan comes from a blog post co-authored by him and Chief Scientist Jakub Pachocki on the OpenAI website, titled "Built to Benefit Everyone: Our Plan."
https://openai.com/index/built-to-benefit-everyone-our-plan/
This latest roadmap redefines what OpenAI plans to accomplish over the coming years.
It summarizes the three stages of OpenAI's development.
Phase one: Conduct research toward AGI; phase two: Turn research into products and observe how people use them; phase three: Make AI accessible to all. This is precisely the stage OpenAI is currently in.
The blog states: "The economy is being reshaped around AI, and the central question has become how to make advanced AI affordable, secure, and easy enough to use so that everyone and every organization can benefit."
At this stage, OpenAI set three goals:
First, create an "automated AI researcher";
Second, accelerate economic growth;
Third, give every person on Earth a "personal AGI."
Three goals converge into one main thread: first, teach AI to conduct research; second, use it to accelerate society as a whole; and third, ensure that the fruits of development reach every individual.
In this new roadmap, OpenAI also specifically mentioned a date: March 2028.
By that time, it believes AI systems may already have completed a significant portion of the research work alongside their own researchers.
The timing of the release is also intriguing.
On the very same day this vision statement was released, OpenAI confidentially submitted its IPO draft documents—a capital move toward the public market, alongside a public vision statement, making it hard to dismiss as mere coincidence.
In fact, OpenAI has long been proficient in this strategy.
When it completed its restructuring last year, it championed the slogan "Built to benefit everyone," and now, with its IPO filing, it has added another roadmap.
At every key step in its business or organizational structure, OpenAI releases a grand narrative.
Setting aside the IPO, OpenAI has set three major goals for its third phase—what these mean for ordinary people is what deserves our greater attention.
Automated AI Researcher
Let AI create AI
"Automated AI Researcher" sounds a bit abstract—what does it do, and how does it relate to us?
This term is not the first time it has appeared in this roadmap.
On the same day OpenAI completed its restructuring at the end of October 2025, Altman and Pachocki discussed this path during a live public stream, later covered by media outlets such as TechCrunch.
This new roadmap formally consolidates the scattered statements made over the past several months into a single guiding document.
During that livestream, Pachocki described the system as being able to independently drive large research projects—not just answering questions or writing code.
In March this year, he was exclusively interviewed by MIT Technology Review, setting this goal as OpenAI’s "north star" for the coming years. He said OpenAI is close to building models that can work continuously and coherently like humans, so that in the future, “you’ll essentially be placing an entire research laboratory inside a data center.”
OpenAI’s Chief Research Officer Mark Chen (left) and Chief Scientist Jakub Pachocki (right), from whom the “automated AI researcher” may emerge.
What is the significance of this?
Previously, humans wrote code and invoked AI. Now, OpenAI wants to reverse this: let AI run experiments on its own, find its own errors, and iterate autonomously to achieve recursive self-improvement.
This idea is not new.
As early as 1966, mathematician I.J. Good predicted that a superintelligent machine could design even better machines, triggering an "intelligence explosion" that would far surpass human intelligence.
Delegating the act of "doing research" to AI is the starting point of this chain. Whoever takes this step first holds the steering wheel of technological advancement.
OpenAI has outlined two phases for this path.
In September 2026, first create a "Research Intern" capable of independently tackling several small projects within a limited timeframe; by March 2028, upgrade to a "Full-time AI Researcher" able to independently drive complete research projects over the long term.
Pachocki highlighted the key point: the difference between interns and full-time researchers isn't intelligence, but "how long a time span they can work independently."
This is not the sci-fi scenario of a superintelligence suddenly awakening and rewriting itself; rather, it’s more like breaking down the research process into stages and automating each one: AI assists humans by writing code, tuning parameters, and filtering options, while humans step back to orchestrate these intelligent agents.
On this path, DeepMind’s AlphaEvolve serves as a ready example: this coding agent that automatically improves algorithms is itself used to optimize the processes that train its own models.
AI improving AI has moved from PowerPoint into the lab—and this is what OpenAI is truly betting on behind the date of March 2028.
From distributing the model to giving AGI to everyone
The second and third goals are more closely related to ordinary people.
Accelerate the economy: OpenAI’s position is to use AI to accelerate science, productivity, and economic growth, while ensuring that the benefits are widely shared. It specifically emphasizes that everyone should receive a meaningful share of the prosperity created by AI.
This was not impulsive.
As early as April this year, OpenAI released "Industrial Policy for the Intelligence Age," outlining its vision for how to "divide the pie" in the AI era and proposing to "put humans first."
In April, OpenAI released an industrial policy proposing to share the prosperity of AI through public wealth funds and shorter workweeks.
Two of the comments were the most specific.
First, establish a Public Wealth Fund, jointly funded by the government and AI companies, with the fund investing in AI growth and distributing returns directly to every citizen as an "AI dividend."
Second, pilot a 32-hour, four-day workweek, converting the efficiency gains from AI into shorter working hours without reducing wages.
Accelerating the economy is primarily a macro-level perspective, while personal AGI brings this macro vision down to every individual.
Personal AGI enables every person on Earth to use an AGI in their own way: to help you understand a medical bill, learn a new skill, start a small business, care for elderly parents, or turn an idea into reality.
This is somewhat like the impact of the electricity revolution on human society.
OpenAI pulls the camera back to 1920s rural America, before electricity arrived—drawing water, washing clothes, preserving food with ice, and ending the day when the sun set.
It was only after electricity arrived that the lives of ordinary people were truly transformed. Likewise, AI should also become as integral to everyone’s daily life as electricity.
Full automation is not our goal.
In the blog, OpenAI also mentioned: Fully automating everything is not the future we want, because such a future would be both empty and dangerous.
It presents a counterintuitive logic: the stronger AI becomes, the more important humans become. In setting direction, making trade-offs, and bearing responsibility, the importance of humans will become increasingly evident—“deciding what is worth doing will, in the long term, remain humanity’s most critical role.”
OpenAI also called for the potential establishment of an international coordination mechanism to slow the development of frontier AI if safety, alignment, and societal resilience fail to keep pace with AI advancements.
As early as 2023, OpenAI suggested that there should be an organization similar to the International Atomic Energy Agency to oversee superintelligence.
Last week, Anthropic also publicly stated that AI progress is happening too quickly, and leading labs may need to slow down or even pause to allow societal structures and alignment research to catch up.
Anthropic’s Marina Favaro and Jack Clark wrote that the world needs to establish a verifiable coordination mechanism to give advanced AI development the option to slow down or temporarily pause when necessary.
AI capabilities are accelerating rapidly, and two laboratories holding the world’s most advanced models have independently placed safety on the agenda.
Setting a brake on cutting-edge AI research is moving from niche discussions to an open agenda at leading labs, gradually becoming an industry consensus.
Releasing a new roadmap at a critical moment for the IPO
Why was this roadmap released at this particular time?
On the same day this article was published, OpenAI confirmed that it had secretly filed for an IPO.
The timing of this release speaks for itself.
According to OpenAI itself, it is currently in the third stage of “making advanced AI accessible to all,” a stage in which it repeatedly emphasizes that the use of advanced AI should be widely distributed. Concentrated power makes society vulnerable, while decentralized power makes society more resilient and free.
In other words: the question of AGI has quietly shifted from “whether it can be built” to “who will own it once it is built.”
In this roadmap, OpenAI did not provide a specific timeline for AGI, but instead focused on AI accessibility and safety. It also noted that the endpoint of AGI is not "full automation of everything," and that humans remain at the core and center of importance.
No one, not even those building it, can agree on when AGI will arrive.
Musk bets on 2026; Anthropic CEO Dario Amodei believes “powerful AI” could arrive as early as 2027; and DeepMind’s Demis Hassabis said in March that it might happen around 2030, when “a system will emerge that deeply understands everything around you.”
Based on the above insights from experts, AGI could emerge in any given year over the next five years.
Returning to OpenAI’s new initiative, if by March 2028 its AI system can collaborate effectively with its own researchers to complete a significant portion of research, the role of humans will still be indispensable.
Perhaps a more important question than when AGI will arrive is whether humans can still hold onto their most fundamental responsibilities and keep steering the wheel when machines begin to conduct research and accelerate the entire world on our behalf.