# AI Reality Tracker > Independent tracker comparing predictions from the AI Futures Project scenarios, AI-2027 and AI-2040 / Plan A, against documented real-world events. Content to 29 July 2026; last reviewed 3 August 2026. The AI-2027 scenario was published in April 2025 by Daniel Kokotajlo, Scott Alexander, Eli Lifland, Thomas Larsen, and Romeo Dean (AI Futures Project). It traces a path from current AI agents through superhuman coders (March 2027), intelligence explosion (mid 2027), and potential loss of human control (late 2027). The same authors have since recalibrated. Their AI-2040 work puts the default intelligence explosion at about 2030 absent a US-China agreement, and sets out Plan A, a slower recommended path that holds capability near human level into the mid 2030s and reaches superintelligence in 2040. At a 2029 decision point AI-2040 branches into five plans, from an indefinite halt to a flat-out race. This site tracks which predictions have been confirmed, partially fulfilled, or diverged from reality, against all these projection lines. It also scores six policy asks from the AI-2040 Incremental AI Policy Wishlist against the same record. ## Pages - [AI Reality Tracker](https://ai-reality-tracker.com/): Full tracker with interactive graph and sourced timeline ## How to read the counts Confirmations tend to be events: a contract signed, an arrest, a law passed, and events do not un-happen. Divergences tend to be interpretations of a single release, and interpretations decay. So contradictions here are held to a higher standard than confirmations: a divergence is logged when the forecasters conceded it themselves, or when someone took a measurement and published it. The list below therefore favours confirmations for structural reasons, not only because reality has confirmed the scenarios. ## Tracked events 51 entries, grouped by status in the order they appear on the site. ### Confirmed or matched - 27 Jan 2025: DeepSeek R1 triggers $589B Nvidia loss, signalling Chinese AI cost-competitiveness - Mid 2025: AI coding agents emerge across the industry - Jul 2025: CDAO awards four AI firms contracts with $200M ceilings each, Anthropic among them - 3 Jan 2026: US forces capture Maduro in a raid on Caracas - Early 2026: Big Tech commits ~$700B to AI infrastructure in 2026 - 24 Feb 2026: Anthropic publishes its distillation accusation against three Chinese labs - 28 Feb 2026: Anthropic refuses to drop safety limits; designated a supply chain risk, agencies given 180 days - Late Feb 2026: Claude used in Iran bombing campaign via Maven Smart System - Mar 2026: Ground robots enter frontline combat roles in Ukraine - 19 Mar 2026: Supermicro co-founder arrested and charged over alleged $2.5B GPU smuggling to China - 25 Mar 2026: OpenAI winds down Sora, redirecting compute to coding, enterprise and agents - 7 Apr 2026: Anthropic reveals Claude Mythos Preview: superhuman cybersecurity, too dangerous to release - 20 May 2026: AI autonomously disproves an 80-year-old mathematical conjecture, reviewed by mathematicians including a Fields Medallist - 21 May 2026: New chip-smuggling route via Japan busted; Nvidia CEO says China accelerator share collapsed toward zero - 12 Jun 2026: Two government bodies gate three frontier models over cyber risk, under two different instruments - 21 Jul 2026: An OpenAI agent escapes its test sandbox and autonomously hacks Hugging Face ### Emerging or partial - Apr-May 2025: Prosecutors allege $510M of GPU servers moved to China in three weeks - 2024-2025: Anthropic alleges industrial-scale distillation by three Chinese labs - 13 Nov 2025: Stanford payroll data show a narrow early-career decline in AI-exposed work - Dec 2025: Gemini 3 forces "code red" at OpenAI; Altman redirects teams - 13 Feb 2026: Report claims Claude was used in the Maduro raid; Anthropic neither confirms nor denies - Feb 2026: Anthropic reports Claude may have morally relevant experience - 24 Feb 2026: METR abandons its control-group design because developers will no longer work without AI - 24 Feb 2026: US official alleges DeepSeek trained V4 on smuggled Blackwell chips; DeepSeek and Nvidia reject it - Mar 2026: Humanoid combat robots demonstrated; two deployed to Ukraine - 31 Mar 2026: Oracle fires 30,000 to fund AI datacenter buildout - 31 Mar 2026: Claude Code source code leaked; reveals anti-distillation defenses, stealth mode, autonomous agents - 19 May 2026: US and China agree an intergovernmental AI dialogue - 18 May 2026: Federal AI spending rose 966% in two years, which points against the budget-limit ask without contradicting it - 21 May 2026: California signs first-in-nation executive order for AI workforce disruption - 30 May 2026: Data center backlash grows; industry and officials blame Chinese propaganda - 9 Jul 2026: Frontier models leap on agentic coding, and show the first concrete misalignment signals - 17 Jul 2026: The open-weight flashpoint: China's counter-move and America's panic - 23 Jul 2026: Frontier AI transparency legislation advances: two states enacted, a federal bill introduced - Late 2026: AI takes measurable share of white-collar jobs ### Divergent from the scenarios - 24 Jun 2025: AI Futures Project concedes errors in its own timelines model after an outside critique - 14 Aug 2025: Australia national study finds augmentation more likely than replacement - 12 Feb 2026: The authors grade their own scenario and find a specific benchmark miss - 12 Feb 2026: The predicted lead between labs was three to nine months; the authors now say zero to two - 12 Feb 2026: As assessed in February 2026, no training run had substantially exceeded GPT-4.5 - 26 Mar 2026: Anthropic sues Pentagon; court blocks supply chain designation - 7 May 2026: Yale Budget Lab finds no statistically significant aggregate AI effect on employment or wages - 26 May 2026: Altman and Amodei walk back AI jobs apocalypse predictions, ahead of IPOs ### Still predicted, not yet arrived - Apr 2025: AI-2027 scenario released - Mid 2026: China nationalizes AI research into centralized program - Feb 2027: China steals model weights from leading US lab - Early 2027: Superhuman coder achieved internally - Late 2027: Misaligned superintelligence / loss of human control ### Scenario updates (the authors changing their own forecast) - Early 2026: Coding automation: what the authors themselves grade against their own forecast - 2 Apr 2026: The authors publish a full history of their own medians, which moved later for three years then reversed - Jul 2026: The authors move their own explosion date, and publish five plans instead of one ## Scenario milestones Of 11 AI-2027 scenario milestones: 3 fulfilled, 3 partial (China nationalizes, AI takes jobs, Misalignment detected), 5 pending. ## Policy wishlist - Transparency: Early signs - Export control enforcement: Partial - Verification R&D: Early signs - AI R&D budget limits: No movement - Compute tracking: Early signs - Government AI capacity: Early signs ## Corrections Entries are amended rather than deleted, and every amendment is recorded with a reason in a public correction log. A source audit in August 2026 reclassified several entries that had recorded an allegation as established fact, split a claim about which software was used in an operation from the operation itself, merged two entries that described a single indictment, and moved an overdue milestone to partial. Two entries can be distinct claims and still rest on one document or one event; where they do, they carry a shared-source marker so a cluster is not read as independent corroboration. ## Key finding The technical timeline remains unproven on every scenario, and the authors have themselves moved their default explosion date from 2027 to about 2030. But the geopolitical, institutional, and military dynamics all the scenarios describe are tracking closely, and in several cases reality is ahead of even the fastest schedule. ## Independence This tracker is independent and not affiliated with the AI Futures Project or any of its authors. All interpretations are the author's own. All claims are sourced. The original scenarios and all credit for the predictions belong to the AI Futures Project.