A hypothetical system whose cognitive performance greatly exceeds the best human minds in nearly every domain. No system has reached it, and there is no agreed test for when one would.
Superintelligence (SI) is an intellect that outperforms the best humans across nearly every field of thought, including science, strategy and social reasoning. It sits one step past AGI, which targets human-level performance across the full range of tasks. No existing system meets either bar. Superintelligence remains a goal: AI labs name it in their mission statements, and researchers study it as a future possibility.
Some systems already beat the best humans in one domain. Deep Blue defeated Garry Kasparov at chess in 1997, AlphaGo defeated Lee Sedol at Go in 2016, and AlphaFold predicted protein structures at near-experimental accuracy in 2020. Each of these systems can do nothing outside its own domain. To count as superintelligence, a system must have that kind of lead in nearly every domain at the same time.
In July 2023 OpenAI started a "Superalignment" team to control systems smarter than humans, and it committed 20% of its compute to that team. The team was dissolved in May 2024. In June 2024 Ilya Sutskever co-founded Safe Superintelligence Inc., whose stated goal is to build superintelligence as its only product. In June 2025 Meta grouped its AI research under Meta Superintelligence Labs. Each of these is a stated goal. No lab has claimed a working system.
Current AI visibility work does not depend on the concept. Brands are cited and recommended by large language models with retrieval attached, and by AI agents. These systems are narrow, and superintelligence describes a destination that some labs have announced.
