How AI companies define superintelligence

AI companies do not use “superintelligence” the same way. OpenAI, Meta, Safe Superintelligence Inc. and Microsoft name it as a goal; Anthropic and Google DeepMind mostly speak of “powerful AI” and “AGI.” Microsoft’s version is bounded, Meta’s is personal, and some firms use the word without defining it. A company definition is a statement of intent, not a measurement.

Why does it matter how companies define the word?

Company definitions matter because the companies are the ones building the systems, and their wording sets public expectations.

The classic definition, from philosopher Nick Bostrom, is an intellect that greatly exceeds human performance in virtually all domains. It is explained in the definition guide. Several companies now use the same word for something smaller, or for a direction rather than a destination.

The same word on two company websites can describe two different things. Below, each company is quoted briefly, with the date and a link to the original.

What does OpenAI mean by superintelligence?

OpenAI’s closest thing to a definition of superintelligence dates from May 2023, when it described superintelligence as “future AI systems dramatically more capable than even AGI.” Its founding document defines AGI, not superintelligence. The 2018 Charter describes AGI as “highly autonomous systems that outperform humans at most economically valuable work.”

The company began writing about superintelligence directly in 2023.

• May 22, 2023, “Governance of superintelligence.” OpenAI wrote that it was conceivable that within the next ten years, AI systems would exceed expert skill level in most domains. The post proposed coordination among leading developers and an international oversight body for the most capable systems.
• July 5, 2023, “Introducing Superalignment.” OpenAI said it believed superintelligence “could arrive this decade” and committed 20% of the compute it had secured to date to the Superalignment effort, with a goal of solving the core technical problem within four years. In May 2024, Fortune reported, citing unnamed sources, that the team never received that compute and had been disbanded.
• January 2025, “Reflections.” Sam Altman wrote that OpenAI was confident it knew how to build AGI as traditionally understood, and was turning its aim to “superintelligence in the true sense of the word.”
• June 10, 2025, “The Gentle Singularity.” Altman wrote that humanity is close to building “digital superintelligence.”
• November 2025, “AI Progress and Recommendations.” OpenAI said it expected AI capable of very small discoveries in 2026 and more significant discoveries in 2028 and beyond. It also said no one should deploy superintelligent systems without being able to robustly align and control them.

OpenAI has also tied its goals to a research milestone rather than a definition. In an October 2025 livestream, Altman set targets of an intern-level AI research assistant by September 2026 and an automated AI researcher by March 2028. On September 6, 2026, OpenAI wrote that, “according to our measurements,” it had reached the first target. It defined a research intern as a system that carries out well-defined research tasks under human direction. That milestone was set, defined and graded by OpenAI.

What does Meta mean by “personal superintelligence”?

Meta announced Meta Superintelligence Labs on June 30, 2025, in an internal memo from Mark Zuckerberg. The unit combined Meta’s foundation-model, product and FAIR research teams. At launch it was co-led by chief AI officer Alexandr Wang, formerly of Scale AI, and former GitHub chief executive Nat Friedman.

On July 30, 2025, Zuckerberg published “Personal Superintelligence.” It says that “developing superintelligence is now in sight” and that Meta’s vision is to bring personal superintelligence to everyone.

Meta’s emphasis is on who uses the system, not only on how capable it is. The essay describes an AI that knows a person well and helps them reach their own goals, delivered through everyday devices such as smart glasses. It contrasts this with a centralized superintelligence that automates most valuable work. It does not give a capability test.

By October 2026, the lab had released models under the “Muse” name, beginning with Muse Spark in April 2026.

What does Safe Superintelligence Inc. mean?

Safe Superintelligence Inc. (SSI) was announced on June 19, 2024 by Ilya Sutskever, a co-founder and former chief scientist of OpenAI, with Daniel Gross and Daniel Levy. Sutskever became chief executive in July 2025, when Gross left to join Meta.

SSI’s website is short. It calls building safe superintelligence “the most important technical problem of our time” and describes the company as a “straight-shot SSI lab, with one goal and one product.” It says it plans to advance capabilities as fast as possible while keeping safety ahead.

The site says superintelligence “is within reach” but does not define it or set a date. As of October 2026, it lists no released product. SSI is the clearest case of a company named for the goal itself.

What is Microsoft’s “humanist superintelligence”?

On November 6, 2025, Mustafa Suleyman, chief executive of Microsoft AI, announced the MAI Superintelligence Team in a post titled “Towards Humanist Superintelligence.”

Humanist superintelligence: “incredibly advanced AI capabilities that always work for, in service of, people and humanity more generally.”

— Mustafa Suleyman, Microsoft AI, November 2025

Suleyman’s version is defined partly by what it is not. The post says it is not an unbounded entity with high degrees of autonomy, and it rejects the idea of a race to AGI. It names three application areas: an AI companion for everyone, “medical superintelligence” for expert-level diagnosis, and plentiful clean energy. The post says medical superintelligence will arrive “in the next few years.”

That is a domain-specific use of the word. Under Bostrom’s definition, superintelligence in one field would not count. Under Microsoft’s, it is the starting point.

Fortune reported that the team followed a restructured Microsoft–OpenAI agreement. The earlier deal had kept Microsoft from pursuing its own AGI research. The same article called the rebranding “better marketing than science.”

What does Anthropic say?

Anthropic’s chief executive prefers “powerful AI” to either term.

In his October 2024 essay “Machines of Loving Grace,” Dario Amodei called AGI an imprecise term with science-fiction baggage and hype. He uses “powerful AI” instead, describing a model smarter than a Nobel Prize winner across most relevant fields, able to work autonomously for long periods and to run as many copies. He wrote that it could come as early as 2026.

In a March 2025 submission to the White House Office of Science and Technology Policy, Anthropic said it expected powerful AI systems to emerge in late 2026 or early 2027.

The vocabulary differs. Amodei compares powerful AI with Nobel Prize winners rather than with workers, a higher bar than the economic AGI of OpenAI’s Charter, but he does not call it superintelligence. Readers comparing labs should compare the descriptions, not the labels.

What does Google DeepMind say?

Google DeepMind mostly talks about AGI rather than superintelligence. Its most detailed public framework is the 2023 paper “Levels of AGI.” It treats artificial superintelligence as the top of a ladder: general performance that outperforms 100% of humans. The paper lists that level as not yet achieved. The AGI vs superintelligence guide walks through each level.

In April 2025, the company published a paper on AGI safety and security. The accompanying post defined AGI as AI at least as capable as humans at most cognitive tasks and said it “could be here within the coming years.”

Who else uses the word?

The term has spread beyond the frontier labs, sometimes with no definition attached.

On September 1, 2026, the cybersecurity company CrowdStrike announced a “Cyber Superintelligence Lab” for AI-driven cyber defense and AI safety research. The press release does not define superintelligence.

Used that way, the word works as a brand name. There is nothing to test it against.

How do the company definitions compare?

They differ on four points: how broad the system is, how autonomous it is, whom it is meant to serve, and when it is expected.

• Breadth. OpenAI’s 2023 wording (expert level in most domains) and Anthropic’s powerful AI (most relevant fields) are broad. Microsoft’s starts with one domain, medicine.
• Autonomy. OpenAI’s Charter and Anthropic’s description require autonomous work. Microsoft explicitly rejects high autonomy.
• Who it serves. Meta’s qualifier is personal, Microsoft’s is humanist and SSI’s is safe. None of the three is a measurable test.
• Dates. OpenAI (within ten years of 2023; “this decade”) and Anthropic (late 2026 or early 2027 for powerful AI) have named timeframes. Meta says superintelligence is “in sight,” and SSI’s website says it is “within reach.” Microsoft says medical superintelligence will arrive “in the next few years.” The timelines guide lists the claims in order.

The table puts each company’s term and wording next to the date it was stated.

The concerns these definitions raise, and the public responses to them, are in the debate guide. The glossary defines each company term in one place.