What Is SI? Super Intelligence and the New U.S. AI Term
SI stands for Super Intelligence in the U.S. executive branch’s new terminology for AI, introduced by the September 29, 2026 executive order. In AI research, superintelligence means intellect far beyond the best human minds across almost every field, as described in Nick Bostrom’s definition.
What superintelligence means in AI research
The research concept combines exceptional ability with breadth: Bostrom’s definition includes scientific invention, sound judgment and social understanding, rather than superiority at a single task (Bostrom’s explanation).
It defines a level of intelligence rather than a particular physical design; Bostrom allows for different implementations, including one computer or a network of computers (Definition and implementation).
Broad intelligence and independent action are also separate questions: Google DeepMind researchers’ Levels of AGI paper treats performance, generality and autonomy as distinct dimensions.
SI, AI, AGI and ASI: the differences
The abbreviation alone does not identify a capability level. The federal naming policy and research terminology use different definitions:
| Term | Full name | Meaning in this context |
|---|---|---|
| AI | Artificial intelligence | A broad technology category; the U.S. statutory definition includes systems that make predictions, recommendations or decisions (Statutory definition). |
| AGI | Artificial general intelligence | Broad cognitive capability, commonly discussed in relation to human performance; definitions and thresholds vary (AGI definitions). |
| ASI | Artificial superintelligence | General intelligence with superhuman performance across a wide range of tasks in the DeepMind researchers’ proposed framework (Levels of AGI). |
| SI | Super Intelligence | Under the federal order, the same technology category covered by the referenced AI definition (Federal SI definition). |
The DeepMind framework separates depth, meaning performance on a task, from breadth, meaning the range of tasks covered (Performance and generality).
Its table places AlphaZero and Stockfish in superhuman narrow AI, while reserving ASI for general superhuman capability; this illustrates why exceptional performance in a specialized domain does not settle a claim about general intelligence (The framework’s classification).
The order changes official language and requests a legislative proposal
President Donald Trump signed the naming order on September 29, 2026; the White House says Super Intelligence is intended to reflect advancing capabilities and technologies that expand human creativity (White House fact sheet).
The directive covers executive-branch communications, including websites, reports and policy documents, where law permits; it does not require rewriting earlier regulations, presidential actions, contracts, grants or historical documents (Implementation provisions).
Section 3 uses the existing statutory AI definition for SI and gives the President’s science and technology adviser 60 days to submit proposed legislative language, including possible changes to that definition and related statutory references (Definition and legislative proposal).
The referenced law, 15 U.S.C. § 9401(3), describes machines that pursue human-defined objectives through predictions, recommendations or decisions affecting physical or virtual environments (The statutory text).
That definition describes how systems process inputs, build representations and infer options for information or action, without requiring them to surpass human intelligence (AI definition in § 9401(3)).
Implementation is already visible at NIST: its Super intelligence page says the agency is updating communications under Executive Order 14434 (NIST’s terminology notice).
NIST describes its SI work as including scientific machine learning, language models and agentic AI, with an emphasis on measuring performance, reliability and security (NIST’s research and evaluation activities).
Why AI and SI still appear side by side
Government usage already differs within the United States: California’s September 30, 2026 announcement says the state is retaining the name Artificial Intelligence (California’s announcement).
This difference shows that the federal executive branch’s naming convention has a jurisdictional context, rather than establishing a universal technical vocabulary. Federal scope, California’s separate position.
Does the SI name show that superintelligence has arrived?
The naming order’s definition supplies no requirement for broad superhuman performance; a reference to federal SI therefore does not establish research-defined superintelligence (The order’s definition).
Capability research addresses measurable tasks and evaluation conditions: METR’s March 2025 study tested software and reasoning tasks, comparing agents’ completion rates with the time those tasks took human professionals (METR’s study).
Within that study, task duration strongly predicted success, and agents performed much less reliably on longer tasks than on short ones (The study’s findings).
The study also identifies uncertainty about extending its trends to future performance and other settings; its findings concern the evaluated tasks, rather than measuring all human intellectual abilities (Limitations and extrapolation).
The distinction matters when reading company announcements too: Safe Superintelligence Inc. presents safe superintelligence as its mission and says it approaches safety and capabilities together as engineering and research problems (SSI’s mission statement).
A stated research goal is evidence of what the organization aims to build; the mission statement itself does not provide a broad capability evaluation demonstrating that the goal has been achieved (SSI’s published statement).
Frequently asked questions
What does SI stand for in AI discussions?
SI stands for Super Intelligence in the September 2026 U.S. executive-branch policy (The naming order).
Is SI the same as AGI?
The federal SI label covers the existing AI category (Federal definition). AGI concerns general capability, with research frameworks comparing performance across many tasks (Levels of AGI).
Is superintelligence the same as ASI?
ASI means artificial superintelligence: in the DeepMind researchers’ framework, it is general AI performing beyond humans across a wide task range (ASI classification).
Does superintelligence have to be conscious?
Bostrom’s definition leaves consciousness and subjective experience open; superior intellectual performance does not, within that definition, require an answer about consciousness (Bostrom’s definition).
Does the new SI name apply worldwide?
The announced policy covers the U.S. executive branch (White House fact sheet). California’s separate decision to retain AI already demonstrates differing usage within the United States (California’s announcement).