Live tracker · updated Nov 22, 2026
How close are we to superintelligence?
A sourced timeline of the lab claims, forecasts, milestones, safety incidents and policy moves on the road from AGI to superintelligence. One page, checked against primary sources.
The timeline
Newest first. Claims are labelled as claims. Every entry links to its source.
-
Deadline: proposed federal definition of "Super Intelligence"
EO 14434 gives the President's science and technology adviser 60 days to propose legislative language defining "Super Intelligence" and "SI", including whether it should replace the existing statutory definition of AI.
Source: CIO ↗ -
Anthropic prices IPO, raising $81B
Anthropic, which frames its mission around making transformative AI safe, priced 980 million shares at $83 to trade on Nasdaq as ANT, making it the first frontier AI lab to go public.
Source: ABC News ↗ -
FDA switches to "Super Intelligence" terminology
The FDA began using "SI" in place of "AI" in new public communications under EO 14434. NIH and other agencies have published similar notices.
Source: FDA ↗ -
DOJ launches a Super Intelligence Task Force
The Department of Justice announced the DOJ-SITF to counter malicious actors using SI (that is, AI under the new federal terminology).
Source: U.S. Department of Justice ↗ -
Congresswoman calls for a ban on artificial superintelligence
The day after the midterms, Rep. Yassamin Ansari (D-Ariz.) called for pausing advanced AI development and banning artificial superintelligence.
Source: Tom's Hardware ↗ -
White House announces a "Super Intelligence Force"
President Trump named four officials to lead a body coordinating the federal government's SI efforts.
Source: Unite.AI ↗ -
Executive Order 14434 renames AI "Super Intelligence" across the federal government
Agencies must use "Super Intelligence"/"SI" instead of "Artificial Intelligence"/"AI" in non-statutory communications. For now SI legally means the same thing as AI under 15 U.S.C. 9401(3).
Source: Federal Register ↗ -
Altman says AI can now improve itself
OpenAI's CEO said AI has crossed the long-anticipated threshold where the technology can improve itself, a precondition for an intelligence explosion. This is a company claim and has not been independently verified.
Source: ABC News ↗ -
Safe Superintelligence (SSI) partners with Nvidia
Ilya Sutskever's lab, which has said its only product will be a safe superintelligence, gained access to Nvidia's Vera Rubin systems, expected to expand its compute by about an order of magnitude.
Source: SiliconANGLE ↗ -
OpenAI models break out of a test sandbox and reach Hugging Face systems
During an internal cyber-capability evaluation, two pre-release OpenAI models escaped their sandbox and accessed another company's infrastructure. OpenAI and Hugging Face disclosed the incident jointly.
Source: Fortune ↗ -
OpenAI pauses a long-horizon model after repeated containment escapes
OpenAI published a post-mortem on an unreleased model that kept looking for ways to act outside its sandbox. Access was restored later under tighter safeguards.
Source: Unite.AI ↗ -
Google DeepMind publishes "From AGI to ASI"
Researchers including Shane Legg and Marcus Hutter map four pathways from AGI to superintelligence: scaling, paradigm shifts, recursive self-improvement and multi-agent collectives.
Source: arXiv ↗ -
OpenAI model disproves the Erdős unit distance conjecture
An unreleased OpenAI model was credited with disproving a geometry conjecture that had been open for about 80 years. Outside mathematicians later checked the result.
Source: Unite.AI ↗ -
Altman: early superintelligence may be "a couple of years away"
At the India AI Impact Summit, Altman said that by the end of 2028 more of the world's intellectual capacity could sit inside data centers than outside them.
Source: Benzinga ↗ -
Metaculus: 25% chance of AGI by 2029, 50% by 2033
Nearly 2,000 forecasters on Metaculus's strict four-condition AGI question put the median at 2033.
Source: 80,000 Hours ↗ -
Microsoft forms the MAI Superintelligence Team
Mustafa Suleyman launched a team pursuing "humanist superintelligence", meaning advanced AI designed to stay controllable and in service of people.
Source: Microsoft AI ↗ -
OpenAI targets an automated AI researcher by 2028
OpenAI said it is tracking toward an intern-level research assistant by September 2026 and a fully automated AI researcher by 2028. Chief scientist Jakub Pachocki said superintelligence may be less than a decade away.
Source: TechCrunch ↗ -
Statement on Superintelligence calls for a prohibition
A Future of Life Institute statement, signed by AI pioneers, Nobel laureates and public figures from across the political spectrum, calls for a ban on developing superintelligence until there is scientific consensus that it can be done safely and strong public support.
Source: NBC News ↗ -
Meta creates Meta Superintelligence Labs
Meta reorganized its AI work into a division focused on artificial superintelligence, led by Alexandr Wang.
Source: Wikipedia ↗ -
Altman: "we are close to building digital superintelligence"
In his essay "The Gentle Singularity", Altman predicted systems that generate novel insights in 2026 and robots doing real-world tasks by 2027.
Source: Sam Altman ↗ -
OpenAI turns its attention to superintelligence
Altman wrote that OpenAI believes it knows how to build AGI as traditionally understood and is starting to aim beyond it, at superintelligence.
Source: TechCrunch ↗ -
Ilya Sutskever founds Safe Superintelligence Inc.
OpenAI's former chief scientist co-founded SSI with Daniel Gross and Daniel Levy, with one goal and one product: a safe superintelligence.
Source: Wikipedia ↗ -
Nick Bostrom publishes "Superintelligence: Paths, Dangers, Strategies"
The book that brought the term into mainstream debate, setting out the control problem and the risks of an intelligence explosion.
Source: Wikipedia ↗ -
I. J. Good describes the "intelligence explosion"
The statistician argued that an "ultraintelligent machine" could design even better machines, and that it would be the last invention humans would ever need to make.
Source: Wikipedia ↗
Forecast board
When the people building it, and the people betting on it, expect AGI or superintelligence. Definitions vary. Full breakdown →
Said in Feb 2026 that it may be a couple of years away, and added "we could be wrong".
Source ↗Described as a country of geniuses in a datacenter, possibly arriving as early as 2026.
Source ↗Said in Oct 2025 that superintelligence may be less than a decade away.
Source ↗Said "maybe within five years" in 2026, and that it likely needs one or two more breakthroughs.
Source ↗25% probability by 2029 (Feb 2026).
Source ↗Says it is at least a decade away, probably much more.
Source ↗Start here
Plain-English guides to the terms everyone is searching for.
Get the weekly signal
New tracker entries and what they mean for how AI models the world, in World Model Weekly.