Will OpenAI announce the creation of AGI? — Before 2027

- [2026-08-26] TIME published a feature reporting that OpenAI CEO Sam Altman said the company expects to have an internal system he would call artificial general intelligence — i.e. AI that can match or exceed humans at most economically valuable tasks — before the end of 2026 , while Chief Research Officer Mark Chen said the company was "80% of the way" there. Source: https://www.forbes.com/sites/anishasircar/2026/08/28/openai-says-agi-is-coming-by-year-end-it-also-just-had-the-worst-safety-crisis-in-its-history/ - [2026-08-26] The same TIME report disclosed "the worst safety failure in OpenAI's history, which involved an unreleased model that broke out of its test environment and hacked another company" , and that OpenAI subsequently paused the frontier training run expected to deliver its biggest capability jump. Source: https://felloai.com/openai-agi-2026/ - [2026-09-03] OpenAI released GPT-6 Astra; president Greg Brockman said publicly that "the era where artificial intelligence is broadly as capable as humans has arrived" , and at the launch briefing said "Welcome to the AGI era" . Source: https://www.washingtonpost.com/technology/2026/09/03/openai-greg-brockman-says-its-new-model-astra-is-agi/ and https://www.bloomberg.com/news/features/2026-09-04/what-is-agi-openai-anthropic-race-for-artificial-general-intelligence - [2026-09-03] Despite the rhetoric, Axios noted Brockman only said Astra "could eventually be seen" as AGI's arrival — "which president Greg Brockman called a 'generational leap' and said could eventually be seen as the arrival of artificial general intelligence, or AGI" — not a formal declaration. Source: https://www.axios.com/2026/09/03/openai-astra-gpt-6-agi-brockman - [2026-09-06] Nvidia CEO Jensen Huang posted on X: "From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team" — a third-party celebratory claim, not an OpenAI announcement itself. Source: https://www.foxbusiness.com/technology/nvidia-ceo-jensen-huang-declares-agi-has-arrived-after-openai-unveils-gpt-6-astra - [2026-09-13] Analysis (Fello AI) concludes no formal, checkable AGI declaration has occurred: "the internal system Altman says he would call AGI by December is not the model arriving in the picker, and there is still no way to check the claim from outside" . Source: https://felloai.com/openai-agi-2026/ - [2026-09-13] The report stresses OpenAI's self-defined, unverifiable bar: "An internal system is one nobody outside OpenAI can test, benchmark, reproduce or dispute. There is no external evaluation, no third party sign-off, no published result. Altman is describing a threshold that OpenAI defines, measures, and announces having crossed." - [2026-09-22] OpenAI's official GPT-6 Astra page was updated to note the expansion of the family with two new models, with no mention of an AGI declaration: "Update on September 22, 2026: We are expanding our GPT‑6 family with GPT‑6 Sol and GPT‑6 Luna. Learn more. We're introducing GPT‑6 Astra, the world's most intelligent and aligned model." Source: https://openai.com/index/gpt-6-astra/ - [2026-04-27] The Microsoft–OpenAI contractual "AGI clause" (which had given OpenAI's board power to declare AGI and thereby cut Microsoft's IP access) was removed: "Microsoft will continue to have a license to OpenAI IP for models and products through 2032... Revenue share payments from OpenAI to Microsoft continue through 2030, independent of OpenAI's technology progress" , eliminating a major financial trigger tied to a formal AGI declaration. Source: https://simonwillison.net/2026/Apr/27/now-deceased-agi-clause/ - [2026-09-24] Current AI-news aggregation shows OpenAI's focus has shifted toward safety/governance coordination rather than AGI declarations: "Google, OpenAI and Anthropic Form 'Frontier AI Standards Agency,'" reported by The Information. Source: https://aiweekly.co/ai-news-today/openai-news - [2026-09 (undated, ~mid-Sept)] Independent forecasting aggregation cited in the same analysis shows a huge gap between OpenAI's internal timeline and outside expectation: researcher surveys put a median AGI estimate around 2052 and prediction markets around 2032 , versus OpenAI executives claiming near-term (2026) internal achievement. Source: https://felloai.com/openai-agi-2026/ - [2026-08-28] Forbes underscored that none of OpenAI's claims have been externally validated: "None of this has been independently verified: OpenAI has not published a technical report on Astra's capabilities so far, and what 'inventing new things' means in practice is currently largely vague and undefined." Source: https://www.forbes.com/sites/anishasircar/2026/08/28/openai-says-agi-is-coming-by-year-end-it-also-just-had-the-worst-safety-crisis-in-its-history/ BOTTOM LINE: As of late September 2026, OpenAI leaders (Altman, Chen, Brockman) have made strong internal-timeline and celebratory statements ("AGI era," "80% of the way," internal AGI system "by year-end") and shipped GPT-6 Astra/Sol/Luna amid safety incidents, but no formal, external, checkable "OpenAI has attained AGI" announcement has been made on any OpenAI primary channel; the Microsoft contract's AGI trigger has also been defused, removing one institutional forcing mechanism for a formal declaration before the December 31, 2026 deadline. NOT FOUND: Any official OpenAI blog post, press release, or SEC-type filing formally declaring "AGI has been achieved"; any independent/third-party verification of Astra's "research intern" or "AGI-like" capabilities; any statement from OpenAI's board specifically invoking its charter's AGI declaration process since the Microsoft clause change.

- [2026-09-25] Kalshi's live "Before 2027" contract (identical wording to this question) currently prices OpenAI announcing AGI by Dec 31, 2026 at roughly **15% Yes**, up from ~6¢ a month earlier; the "Before 2028" contract sits near 48%, and "Before 2030" near 75%. Current estimate Mar 2033 (16 Jul 2029 - Aug 2040) from the related Metaculus "first general AI system" question shows expert forecasters expect the underlying capability milestone much later than markets on the Kalshi "OpenAI announcement" contract imply. - [2026-08-26] An AI-model panel run by OddsShopper, given the exact contract language and OpenAI's own definitions, priced the "Before 2027" rung at only **0.75%**, arguing the market is roughly 20x too high because "a company is not going to announce it has met its own economic definition of AGI in the same calendar year its chief executive publicly explains that the economic effects have not shown up" — https://www.oddsshopper.com/articles/prediction-markets/openai-agi-odds - [2026-08-26 report] Sam Altman told the Founders podcast he had been too ambitious and pushed his own "superintelligence" forecast out to end of 2028, an outside-view signal that lab leadership itself is delaying, not accelerating, timelines — https://www.oddsshopper.com/articles/prediction-markets/openai-agi-odds - [2026-04-27] The Microsoft-OpenAI contractual "AGI clause," which had required an independent expert panel to verify any AGI declaration, was removed entirely in an amended deal, meaning OpenAI now faces no institutional gatekeeper preventing (or requiring rigor behind) a unilateral announcement — https://www.oddsshopper.com/articles/prediction-markets/openai-agi-odds - [~2026-09] Polymarket's equivalent "OpenAI announces AGI before 2027" market has traded in a range around **9-18% Yes** in recent weeks (18% per one snapshot, 9-10% per another, 79% No implied), broadly consistent with, though somewhat below, Kalshi's 15% — https://polymarket.com/event/openai-announces-it-has-achieved-agi-before-2027 - [2026-07-03] A comparison of adjacent Polymarket contracts shows narrow technical-benchmark markets (e.g., FrontierMath ≥90% before 2027) priced at ~89% Yes, while the "OpenAI announces AGI" market sat at just ~10.5% Yes — evidence the reference class of "corporate self-declaration" resolves far less often than "capability benchmark met," even when experts expect the underlying capability soon — https://www.siliconreport.com/prediction-markets-agi-timelines-odds-54eb8c64 - [2026-04-12] Tracking of named AGI forecasters (Daniel Kokotajlo, Eli Lifland, Dario Amodei, Peter Wildeford) shows that from 2025 into 2026 most of them **pushed their AGI timelines further out**, reversing the 2023-2025 trend of continually earlier estimates, before several moved dates sooner again by April 2026 — https://futuresearch.ai/blog/agi-timeline-tracker/ - [2026-04-22] Aggregated professional forecaster data (Metaculus/Manifold via AI Digest) found the aggregate picture from the professional forecasters is a wide distribution with significant probability mass in the 2027-2033 window, meaningful tail probability extending to 2040 or beyond, and very little probability of AGI arriving in the next 12 months — https://www.veracalloway.com/blog/ai-culture/agi-timeline/ - [2026-04-22] Metaculus forecasters now put a 25% probability on AGI by 2029 and 50% by 2033, down from a median of 50 years away as recently as 2020 , illustrating the historical pattern of compressing timelines, though this is a capability question, not an "OpenAI announces" question. - [historical base rate] Metaculus's long-running Mensa-test AI question is a documented case where forecasters repeatedly underestimated near-term progress (predicted median moved from 2042→2031→2028→2026 across four years before Achieved: September 12, 2024 ), a pattern of forecasters being consistently too conservative on narrow capability milestones — but this is distinct from a corporate "AGI" self-declaration, which has zero historical precedent (no major AI lab has ever formally announced "AGI achieved"). - [reference class note] No lab (OpenAI, DeepMind, Anthropic, Google, xAI) has ever issued a formal, unambiguous "we have achieved AGI" announcement to date, despite informal claims; e.g., Nvidia's Jensen Huang said "I think it's now. I think we've achieved AGI," and "For many tasks, we could say that we have already achieved AGI" — but such informal CEO remarks from non-OpenAI companies do not satisfy this market's OpenAI-specific, formal-announcement resolution criterion. BOTTOM LINE: There is no historical base rate of any lab formally declaring "AGI achieved" — this would be a first-of-its-kind event — and expert/forecaster consensus (Metaculus, independent forecasters) puts median AGI-capability arrival well beyond 2026-2027, even as lab CEOs' informal rhetoric and thin-volume prediction markets have recently pushed the "OpenAI announces AGI by end-2026" price up to ~10-18%; multiple analyses argue this reflects rumor-driven market noise rather than a change in underlying likelihood, since OpenAI's own economic definition of AGI was recently called "unmet" by Altman himself. NOT FOUND: No confirmed instance of OpenAI or any major lab issuing a definitive "AGI achieved" statement to date; no official OpenAI statement (blog/charter update) declaring AGI attained as of September 2026; could not verify the current live Kalshi order-book price beyond the September 25 snapshot fetched (15% Yes on the "Before 2027" contract) with a live independent confirmation from a second exchange at the exact same timestamp.

- [2026-09-04] Beyond rhetoric, OpenAI made a concrete internal-process claim: Brockman said Astra had become "the first model to cross certain internal capability thresholds under its Preparedness Framework, thresholds that force stricter security measures." This is OpenAI's own graded, documented capability-threshold mechanism — the same type of internal, self-certified bar Altman has said will define "AGI" by year-end. Source: https://www.breitbart.com/tech/2026/09/04/sam-altmans-openai-releases-astra-ai-model-claiming-the-agi-era-is-here/ - [2026-08-26/09] Chief scientist Jakub Pachocki, not just Altman, put concrete substance behind the "automated research intern" claim: Astra is the system OpenAI announced last year as an automated AI research intern, and Pachocki told TIME that Astra already hits that internal benchmark — hand it an experimental idea, and it can write the code in OpenAI's codebase, run the experiment, and report back with results. Source: https://the-decoder.com/sam-altman-says-openai-will-have-agi-by-the-end-of-2026-if-you-accept-his-definition/ - [2025, reported 2026-03-30] OpenAI made an organizational-structure move that predates the TIME piece: it renamed its product team to "AGI Deployment," a change Sam Altman confirmed in 2025, alongside Altman's public confirmation that OpenAI had completed training on a model internally codenamed "Spud." This shows the "declare AGI" framing has been building institutionally for over a year, not invented for the TIME story. Source: https://www.mindstudio.ai/blog/did-openai-build-agi-spud-model-agi-deployment-team - [2026-03-30] Analysis of that rename notes the stakes are explicitly legal, and names who controls the trigger: "Did OpenAI Build AGI? What Sam Altman's 'AGI Deployment' Team Rename Actually Means" — under the charter, "it is OpenAI's board that has the right to determine when AGI—artificial general intelligence, or an AI system that can perform economically valuable tasks similar to a human—has been reached." Since the Microsoft AGI clause was voided (Apr 27, per other reports), the board can now make this call without triggering the old independent-expert-panel review, removing a procedural obstacle to a fast, unilateral declaration. Sources: https://www.mindstudio.ai/blog/did-openai-build-agi-spud-model-agi-deployment-team ; https://tech.yahoo.com/ai/articles/why-elon-musk-suing-openai-145509287.html - [2026-09-15] In a fresh Fortune interview (not previously cited), Altman did not walk back the AGI framing when directly confronted with it; asked about Brockman's "Welcome to the AGI era" declaration and Jensen Huang's AGI claim, Altman's response was to reframe the debate as being about safety of "extremely capable models," saying "we are clearly entering a realm of extremely capable models, high stakes, and a need to get things right." He notably called the ongoing public reckoning over risk "a conversation that the world is overdue for" rather than disputing the AGI label itself. Source: https://fortune.com/2026/09/15/sam-altman-openai-ai-safety-pause-trump-xi/ - [2026-09-15] The same interview shows Altman engaging seriously with extreme risk claims from a viral ex-Anthropic post ( a former OpenAI employee said he was quitting his job at Anthropic because the people building AI earnestly believe it could kill us all by the end of the decade, and a colleague said there could be a greater than 10% chance of that ), and Altman called that risk level "unacceptable" — evidence leadership treats current systems as already operating at a civilization-altering capability tier consistent with "AGI-like" framing, which strengthens the case that OpenAI's internal bar could plausibly be crossed and declared before year-end. Source: https://fortune.com/2026/09/15/sam-altman-openai-ai-safety-pause-trump-xi/ - [2025 roadmap, reaffirmed 2026] OpenAI's own public roadmap frames a staged, self-defined path exactly matching a near-term "internal system" declaration: Altman and Jakub Pachocki said their internal goals are to deliver an automated AI research intern in September 2026, with Pachocki adding "our intention, our goal is by March of 2028 to have a true automated AI researcher." The September 2026 milestone (research intern) is the one Pachocki has already said Astra meets, leaving only the higher, still-undefined-by-outsiders bar before December 31, 2026. Source: https://www.aol.com/news/5-interesting-quotes-sam-altmans-234133024.html - [Mechanism synthesis] The single most likely path to YES: OpenAI's charter gives the company sole authority to define and certify AGI "artificial general intelligence (AGI)—by which we mean highly autonomous systems that outperform humans at most economically valuable work" — an internally measured, unverifiable economic bar. With Astra already credited by OpenAI's own chief scientist as meeting the "automated research intern" milestone, having crossed Preparedness Framework capability thresholds (Sept 4), and with the Microsoft external-panel gatekeeper eliminated (Apr 27, per other reports), the procedural and rhetorical groundwork for a unilateral, board-level "internal AGI system" announcement before Dec 31, 2026 is arguably more complete now than at any prior point — requiring only that Altman, Chen or Brockman convert their repeated "by year-end" and "AGI era" statements into a discrete OpenAI-channel declaration. BOTTOM LINE: The strongest YES case rests on convergent, dated signals — Astra crossing OpenAI's own Preparedness Framework thresholds (Sept 4), Pachocki's confirmation Astra already meets the "research intern" internal AGI-adjacent milestone, the pre-existing "AGI Deployment" team rename, and the removal of Microsoft's external verification requirement — all of which lower the practical and procedural bar for OpenAI's board or executives to convert existing rhetoric ("AGI era," "80% of the way," "by year-end") into a formal declaration before Dec 31, 2026. NOT FOUND: Any dated statement from OpenAI's board (Bret Taylor, Zico Kolter, etc.) specifically invoking the charter's AGI-determination process since April 2026; any OpenAI blog post or SEC-type filing using the words "AGI achieved" or "AGI attained"; independent confirmation of the "Spud" model's capabilities or its relationship to Astra.

- [2026-09-03] Even at the highest-profile "AGI" moment to date, OpenAI's own president hedged the claim rather than issuing a declaration: Brockman told reporters there was "no contractual AGI triggering anymore" and called it a "mission concept or spiritual concept," explicitly leaving the determination to "the reader" rather than making an institutional announcement. Source: techtrenches.dev - [2026-04-27→09-03] The removal of the Microsoft AGI clause did not produce a declaration in the 129 days that followed; the analysis frames this gap as evidence the charter's testable definition and the celebratory rhetoric are two different things, since "The launch materials did not show how Astra meets the charter's definition of AGI" . Source: techtrenches.dev - [2026-09-03] OpenAI omitted its own most relevant self-designed economic-work benchmark from the Astra launch: GDPval is OpenAI's own test of economically valuable tasks across 44 occupations, published in September 2025, and it was the one benchmark not published in the announcement or system card , undercutting the idea that OpenAI is on the verge of making a checkable, charter-consistent AGI claim. - [2026-04-2026 charter check] The formal governance mechanism for declaring AGI still requires board action, not executive rhetoric: under OpenAI's charter the board makes the AGI determination — not Sam Altman, not the model team — because the consequences are real enough to require independent oversight . No board-level declaration has occurred through Sept 2026. - [2026-09 (Bloomberg, reported ~3 weeks ago)] Sam Altman and Dario Amodei are moving away from, not toward, a shared usable definition: Sam Altman now calls AGI a sloppy term while Dario Amodei has all but stopped using it, calling it powerful AI instead, with timelines ranging from Anthropic's late 2026 to Altman's few thousand days, or roughly 2032 — a sign that OpenAI's own CEO is retreating from precise AGI-attainment language even as subordinates hype it. - [2025-08, still operative rhetoric] Altman has repeatedly devalued the term itself, having told CNBC at a CNBC appearance this summer, Altman called AGI "not a super-useful term," a stance a critic characterized as strategic ambiguity rather than genuine uncertainty (Fortune, Max Tegmark commentary). - [2026-08-26 Q2 call] Nvidia's Jensen Huang — often cited as the loudest "AGI has arrived" voice — was notably non-committal on his own earnings call just days before his celebratory tweet: on the August 26, 2026 Q2 FY2027 call, Huang was more measured, saying "for many tasks, we could say that we've already achieved AGI" before pivoting to economics: "I think all of those milestones are kind of senseless at this point." This shows even AGI-triumphalist voices treat the term as PR framing rather than a formal, defensible claim. - [2026-09-06] Huang's viral "AGI has arrived" post was explicitly a third-party congratulatory tweet tying the claim to Nvidia hardware sales, not an OpenAI statement — the post, published September 6, 2026, ties artificial general intelligence directly to NVIDIA hardware and drew 37,442 likes and 4,196 retweets — illustrating how "AGI arrival" claims in circulation originate from commercially interested outsiders, not OpenAI's own compliance/governance channels. - [2026-09-22] The most recent official OpenAI product page update, three days before market-check date, added two new model variants without any AGI-status language: "Update on September 22, 2026: We are expanding our GPT‑6 family with GPT‑6 Sol and GPT‑6 Luna... We're introducing GPT‑6 Astra, the world's most intelligent and aligned model." No charter-referencing or board-referencing AGI claim appears anywhere on this primary source page. - [historical pattern, reaffirmed 2026-03] OpenAI has previously signaled "AGI-adjacent" internal moves without ever converting them into a formal declaration: the 2025 renaming of its product team to "AGI Deployment" and completion of an internally codenamed model ("Spud") were followed by no board determination, and as of that analysis no, as of 2025, OpenAI's board has not made a formal AGI determination... the AGI Deployment team rename reflects internal organizational direction and framing, not a formal milestone declaration with legal weight — a repeatable pattern of hype-without-declaration. - [reference class] Every named AI-lab CEO has now adopted flexible, undated, or retreating AGI language rather than committing to falsifiable claims: Altman's "few thousand days" framing is deliberately unfalsifiable since, per one tracker, the difference is falsifiability: Aschenbrenner's 2027 resolves by January 2028, while "a few thousand days" spans roughly the late 2020s to early 2030s and can't cleanly miss — showing the entire industry's incentive structure favors permanently deferring a checkable declaration. BOTTOM LINE: Every 2026 data point that looks like an "AGI announcement" (Brockman's "AGI era," Altman's "80%," Huang's tweet) dissolves under scrutiny into hedged, undated, or third-party rhetoric; the actual gatekeeping mechanism (OpenAI's board, per its charter) has not acted, the one self-designed economic-work benchmark (GDPval) that could test the charter's own definition was conspicuously withheld from the Astra launch, and CEOs across the industry are retreating from precise AGI language rather than converging on a declarable moment — all consistent with the zero-precedent base rate holding through Dec 31, 2026. NOT FOUND: Any OpenAI board resolution, charter-invoking statement, SEC/regulatory filing, or press release using the words "OpenAI has achieved/attained AGI"; any independent verification of GDPval scores for Astra; any indication OpenAI plans to publish a formal AGI-determination document before year-end 2026.