Artificial superintelligence (ASI) means hypothetical AI that substantially surpasses leading human experts across many intellectual domains—not simply human-level AGI. No arrival date is independently verified. Cortist’s provisional, unscored F-002 estimate is a 24% chance of independently verified ASI by 31 December 2040.
Research-based topic hub · Editorial estimates are provisional, not independently verified outcomes. · Public F-002 v0.1 JSON · Forecast Protocol · 11 dated source statements
Model v0.1 working judgment, October 22, 2026. Not a statistical measurement and not yet eligible for scoring: resolution criteria must be locked first.
Cumulative chance by year-end · Subjective model v0.1 · Method ↗
Start with the definition and evidence threshold. Explore the provisional forecast, original record and dated research below.
ASI refers to hypothetical AI capability far above leading human experts across many intellectual domains. It is not equivalent to a chatbot or a system that tops one benchmark.
Read the ASI reference guide →There is no verified arrival date. Cortist’s F-002 v0.1 assigns a provisional, subjective 24% chance of independently verified ASI by 31 December 2040. The criteria are not locked and the forecast is not scored.
Inspect forecast F-002 →AGI concerns broad transferable competence across tasks. ASI goes substantially beyond the strongest human experts in many domains. Evidence for one would not automatically prove the other.
Compare with AGI →Independent evaluators would need transparent, repeatable evidence of sustained capability far above human expertise across diverse domains—not a single vendor demonstration. There is no universal resolution standard.
Read the verification protocol →Guide clarified 3 November 2026 · Forecast v0.1 recorded 22 October 2026, unchanged · How Cortist evaluates evidence
Future probability changes require a new dated version and an evidence-based explanation; old records remain available.
One recorded draft (v0.1). No newer probability has been published. The original stays available.
Expert opinions alone do not automatically change a Cortist forecast.Artificial superintelligence (ASI) is a hypothetical AI system whose broad intellectual abilities substantially surpass leading human experts. There is no independently accepted ASI milestone. Cortist’s F-002 is a preliminary, subjective 24% cumulative estimate by the end of 2040, not a measured or scored probability.
Evidence Pulse 2.0 checks the published Cortist News index when JavaScript is available. The source-linked, dated research articles remain accessible through Cortist News.
Latest topic-linked reporting, compared with the separately versioned Cortist assessment. News does not automatically update a forecast.
Evidence interpretation: Later safety and control reporting adds context on increasingly capable AI systems.
What remains unproven: A safety incident or control study cannot establish superintelligence across domains.
What to verify next: Cross-domain expert baselines, evaluation independence and accepted event definition.
Snapshot of 3 linked reports. Context is not independent proof or a measured probability change. Dated evidence index · Original forecast v0.1.
AGI concerns broad, adaptable capability across many tasks, with the exact threshold dependent on a stated definition. ASI adds a much more demanding question: can a system substantially outperform leading humans across strategically important domains and sustain that performance outside curated demonstrations? Superhuman play in a game or superior performance on selected exams is not proof of superintelligence.
A defensible classification would distinguish breadth, peak performance, autonomy and reliability. Google DeepMind’s Levels of AGI helps illustrate why generality and performance should be measured separately. This distinction matters because a forecast about AGI cannot simply be relabeled as a forecast about ASI.
One possibility is incremental progress through larger models, better training, stronger tools and more reliable agent workflows. Another involves AI-assisted research accelerating parts of model development. Other paths could face bottlenecks in energy, data, hardware, evaluation, coordination or real-world deployment. These are conditional scenarios, not a known sequence of events.
Self-improvement is a research question rather than an automatic consequence of achieving AGI. A system that writes code or recommends experiments does not necessarily have the resources, authorization, independent validation or operational continuity needed to rapidly redesign itself. The speed of a transition, if any, therefore remains deeply uncertain.
Evidence would need independent assessment across multiple hard and unfamiliar domains, credible comparisons against expert teams and sustained reproducibility. Evaluate performance on new research problems, not only historical exams. Inspect failures, dependencies on human reviewers, costs, tool access, adversarial resilience and whether results survive replication at a different institution.
The 2026 International AI Safety Report distinguishes observed advances from uncertain future capabilities and describes major disagreement about high-end scenarios. This is a reason to document both acceleration and counterevidence. Governance claims also need their own evidence: passing a benchmark is not proof that a system is controllable, safely deployed or economically transformative.
F-002’s 24% by 2040 refers to the cumulative probability assigned by Cortist’s provisional model v0.1 to its intended ASI event by year-end 2040. The curve is an editorial forecast, not data points collected from reality. Its criteria are not independently locked, and there are no resolved or scored ASI outcomes. The existing year slider remains available for exploring this preliminary view.
New independent evaluations, reproducible scientific breakthroughs, robust capability plateaus or meaningful limitations in deployment could change the assessment. Changes would require a dated, transparent forecast revision while preserving the originally published values and evidence trail. Statements by individual researchers should retain their original definitions, dates and uncertainty.
AGI is about adaptable competence over many types of work. ASI adds a far more demanding threshold: strong and repeatable performance beyond expert human ability across multiple important domains. Excellence in chess, mathematics or programming alone would not establish either broad milestone. A credible comparison must disclose model, tasks, tools and expert baselines.
Google DeepMind’s Levels of AGI separates breadth from depth and treats autonomy as an additional consideration. The framework is a research proposal, not an organization that officially certifies AGI or ASI.
Systems that help write code, design experiments or evaluate models might shorten some development cycles. But recommendations are not automatically valid breakthroughs. Compute budgets, high-quality data, verification, human judgment, physical infrastructure and deployment constraints may remain limiting factors.
Recursive self-improvement is therefore a scenario requiring evidence at every stage, not a known consequence of reaching AGI. Analysts should distinguish measured automation of individual research tasks from confirmed accelerated end-to-end research progress.
Outside evaluations should include genuinely new challenges across science, planning, abstraction and unfamiliar work. Results should be reproducible under transparent rules, with failure cases and assistance reported. A company’s strongest selected demonstration is not enough to establish general superhuman ability.
A claim about ASI also requires a prior event definition. Independent researchers need to know which domains count, how “better than humans” is measured and how sustained reliability is tested. It remains possible to improve selected benchmarks without satisfying those requirements.
F-002 model v0.1 assigns a 24% cumulative subjective chance by 2040. It is not the result of repeated observed ASI events or an independently scored forecast; its criteria are still provisional. The values on the historic year slider are preserved as originally published editorial assumptions.
Predictions from outside experts can use inconsistent definitions of “human-level” or “superintelligent” AI. Cortist records their dates and original wording rather than merging them into a single inevitable timeline. See the forecast record.
No. A capability milestone describes what a system might be able to do, whereas actual harm would depend on behavior, permissions, oversight and deployment. A result showing strong technical ability does not by itself prove loss of control or predict a catastrophe.
The International AI Safety Report 2026 describes uncertainty and disagreement over extreme scenarios. Those questions belong in a separate risk assessment with explicit mechanisms and evidence boundaries.
The Levels of AGI taxonomy illustrates why generality, performance and autonomy need separate definitions. A researcher claiming broad superhuman intelligence would need to show repeatable results across unfamiliar domains and comparisons with suitable experts, rather than highlight a single impressive benchmark.
The International AI Safety Report 2026 documents uncertainty and disagreement about the trajectory of advanced systems. Cortist’s F-002 24% by 2040 is a provisional cumulative judgment. A capability claim also does not independently prove a loss-of-control pathway.
Use each original source within its stated scope. An expert forecast is not evidence of a resolved event.
Source-based reporting on governance, capabilities and verification. News informs the record; it does not automatically settle an ASI forecast.
Ten voluntary safety practices put agent control on the agenda. But the US policy term "super intelligence" is broader than research-level artificial superintelligence.
Read the Cortist analysis ↗We distinguish academic definitions and scenario analyses from precise probability forecasts. Dates and definitions matter.
DeepMind discusses artificial general superintelligence as cognitively more capable than large organizations of humans, exploring scaling, AI paradigm shifts, recursive improvement and multi-agent collectives. The paper does not establish a fixed arrival date.
A foundational definition emphasizes performance far beyond the best human brains across virtually all significant cognitive domains. This is much stronger than beating humans at chess, coding or one benchmark.
Research discussion continues to distinguish superintelligence as a hypothetical capability level from broader uses of the phrase in politics and marketing. The lack of a universal test is a central obstacle to a scored forecast.
11 dated source records, 4 years. Direct original references; no new news articles.
A chronological research and public-statement archive, not a leaderboard. Every entry identifies the speaker, publication date, location, core claim and what the statement does not prove. Summaries are editorial paraphrases, not invented direct quotes.
US agencies published voluntary practices covering preparation, detection, response, recovery and reporting during advanced-agent training and evaluation.
Context: US government policy calls this super intelligence; it is NOT proof of research-level ASI.
Read original source ↗Pachocki reflected on reasoning-model progress and wrote that researchers should take seriously the prospect of machines meaningfully smarter than ourselves within their lifetimes.
Context: Personal long-horizon assessment; no ASI arrival date or independently demonstrated ASI.
Read original source ↗The company reported a strategic partnership that it said would increase its available compute capacity by about tenfold.
Context: Capacity investment toward a goal is not a demonstration of superintelligence.
Read original source ↗Researchers surveyed scaling, paradigm shifts, recursive improvement and large-scale multi-agent collectives as possible routes beyond AGI, stressing major uncertainty and bottlenecks.
Context: Conceptual research; the report supplies no single agreed arrival year.
Read original source ↗OpenAI presented exploratory policy proposals intended for a transition toward superintelligence, emphasizing public participation and institutions rather than only capability milestones.
Context: A policy agenda, not a statement that ASI already exists.
Read original source ↗Zuckerberg argued that personal superintelligence should amplify individual agency and said Meta would direct its research and product efforts toward that future.
Context: Strategic vision and company ambition, not a verified ASI system.
Read original source ↗In an official update, Sutskever said he was leading SSI as CEO and that the company would continue work toward safe superintelligence.
Context: Company mission and leadership update, with no demonstrated ASI milestone.
Read original source ↗Altman wrote that OpenAI was beginning to aim beyond its traditional AGI framing toward superintelligence and the prospect of much faster scientific discovery.
Context: Executive ambition; no verifiable timeline or proof of superintelligence.
Read original source ↗SSI presented itself as a laboratory focused on simultaneously advancing capabilities and safety toward safe superintelligence.
Context: The laboratory launch is documented; a product goal must not be confused with an achieved result.
Read original source ↗In launching its superalignment effort, OpenAI stated its belief that superintelligence could arrive during that decade and described the need for new methods to control systems much smarter than humans.
Context: Historical organizational outlook, conditional and unscored; not evidence of arrival.
Read original source ↗The authors argued that systems surpassing expert skill in most domains within ten years were conceivable and proposed special international coordination and audits.
Context: Scenario analysis rather than a numerical probability or independently resolved ASI forecast.
Read original source ↗Only a publicly reproducible capability demonstration can support the eventual outcome. Lab press releases and benchmark records alone are not sufficient.
Superior results across scientific, reasoning, planning and other broadly relevant domains.
Performance compared with expert human teams, not merely an average person.
Repeatable results on unfamiliar and long-horizon tasks, with limitations exposed.
Public independent assessment under criteria fixed before scoring begins.
These are editorial criteria under development, not an established universal scientific ASI benchmark. See our verification methodology and the F-002 draft record.
No. AGI refers to general human-level capabilities; ASI is a much higher, broadly superhuman threshold. The exact boundaries are debated.
No. It is a subjective model v0.1 estimate of whether an independently verifiable system could exist by the end of 2040. It is neither measured progress nor a locked forecast.
Not by itself. We require clearly defined capabilities, independent evidence and documented resolution rules. Our news coverage provides context, not automatic verdicts.
Compare our AGI watch with superintelligence and see why definition, evidence and time horizon matter.
Explore the evidence on existential risk, loss of control and expert disagreement. The X-001 research question is explicitly unscored.
Read AI Existential Risk ↗News is linked to this research topic, but no article alone proves the associated forecast.
NIST, NSA and CISA set out voluntary safeguards for advanced AI agents. A US government label must not be confused with proof that ASI exists.
Read full news report ↗
New analyses from the UK AI Security Institute and Anthropic describe an AI-agent safety failure—and why stopping an unsafe plan proved difficult.
Read full news report ↗Google DeepMind's leader expects AGI within two to three years. That is an expert timeline, not a measured probability or a verified result.
Read full news report ↗View the Superintelligence news filter or explore the complete seven-question forecast register. The topic classification is editorial and the sources remain dated.
