OpenAI Aims to Build a ‘Legitimate AI Researcher’ by 2028, Says Sam Altman

OpenAI is taking a bold step toward a future where artificial intelligence can conduct independent scientific research. During a recent livestream, CEO Sam Altman announced that the company is tracking toward creating a fully autonomous AI researcher by 2028 — a system capable of carrying out advanced research projects with minimal human guidance.

From Intern-Level AI to Full Research Autonomy

Altman revealed that OpenAI’s deep learning models are rapidly advancing. According to internal progress tracking, the company expects to achieve an intern-level AI research assistant by 2026, setting the stage for the creation of a legitimate AI researcher two years later.

Joining Altman during the session, Jakub Pachocki, OpenAI’s Chief Scientist, explained that this AI researcher will not be a human studying AI, but rather an AI system capable of autonomously performing research tasks. “We’re building systems that can deliver meaningful research outcomes on their own,” he said.

Pachocki described this evolution as a key step toward the era of superintelligence — AI systems that outperform humans across a wide range of critical areas. “We believe deep learning systems could be less than a decade away from achieving superintelligence,” he added.

Shifting to a Public Benefit Corporation

The announcement came as OpenAI completed its transformation into a public benefit corporation (PBC), officially moving away from its original non-profit structure. This transition gives OpenAI greater flexibility to raise capital while maintaining a mission-driven approach.

Under the new structure, the OpenAI Foundation, the company’s non-profit arm, will own 26% of the for-profit entity and guide its scientific direction. The foundation also manages a $25 billion initiative to use AI for curing diseases and advancing global research.

Altman emphasized that this shift will help balance innovation with responsibility. “The PBC model allows us to move faster on our research goals while ensuring that our work benefits humanity,” he said.

Scaling Compute and Algorithmic Innovation

To achieve the ambitious 2028 milestone, OpenAI is betting heavily on algorithmic breakthroughs and scaling up what it calls “test-time compute” — the amount of computational time a model spends thinking through a problem.

Current models can reason effectively for about five hours, performing at levels comparable to top human competitors in events like the International Mathematical Olympiad. Pachocki suggested that extending these models’ reasoning time could enable them to solve far more complex problems.

“In the future, we may dedicate entire data centers’ worth of computing power to a single research question,” he said. This level of computational depth could allow AI systems to make scientific discoveries faster than human researchers, accelerating progress in medicine, physics, and technology.

The Road to Superintelligence

OpenAI’s long-term goal aligns with its broader vision of building safe and beneficial artificial general intelligence (AGI). According to Altman and Pachocki, developing a capable AI researcher is an essential step toward that future.

The company believes that with more compute, data, and refined algorithms, AI can eventually outperform human researchers in discovering new materials, designing medicines, and solving scientific problems currently beyond human comprehension.

While this vision brings excitement, it also raises questions about AI safety and control. Altman stressed that the OpenAI Foundation will oversee safety and ethics as development accelerates. “We want progress, but we also want to do it right,” he noted.

A Massive Infrastructure Commitment

Supporting such advancements requires massive infrastructure investment. Altman confirmed that OpenAI is committing to 30 gigawatts of AI infrastructure, representing a $1.4 trillion investment over the coming years. This infrastructure will provide the computational capacity needed to train and run next-generation models capable of autonomous research.

Altman described this as a necessary foundation for the company’s scientific ambitions. “If we want AI to truly contribute to human knowledge, we have to build systems with the scale and intelligence to do so,” he said.

Looking Ahead

The idea of an AI researcher might once have sounded like science fiction, but OpenAI’s roadmap suggests it could become reality within the next three years. With its new structure, immense infrastructure commitments, and ongoing algorithmic advancements, OpenAI is positioning itself at the forefront of a new era of AI-driven discovery.

Whether this future brings breakthroughs or new challenges, one thing is certain — the race toward creating the world’s first legitimate AI researcher has officially begun.