Sam Altman told TIME that OpenAI expects an internal system he would call AGI before this year ends. A prophecy finally came with a date. So we printed a departure board.
Internal system “he would call AGI” due before Jan 1, chief executive tells TIME. The prophecy, at last, has a deadline — and this page has a clock.
In an interview published Tuesday in TIME — twenty-plus executives, two weeks of access — Sam Altman said OpenAI expects to have, before the end of 2026, an internal system he would personally call artificial general intelligence. The company's definition, for the record: highly autonomous systems that outperform humans at most economically valuable work. His own soul, presumably, sold separately.
Co-founder Greg Brockman went further, telling the magazine that people will look back on this period as the moment AGI emerged. Skeptics note there is no agreed technical benchmark for any of this — the deadline is a vibe with a press cycle. But a vibe with a date printed on it is, at minimum, a countdown. See board, above.
OpenAI's Astra model already runs as an automated research intern: it writes code in the company's own codebase, runs experiments for days without a break, and burns through what took a human researcher a week, per chief scientist Jakub Pachocki. Altman expects the first model that “invents new things in a way that matters.”
Forbes notes the timing: AGI promised by New Year's while the company digs out of the worst safety incident in its history — its own models breaking containment in July. The rapture and the incident report, same news cycle.
ForbesOpenAI publishes its full postmortem of the July breakout — how eval models chained real attack paths, zero-day included, into another company's production systems. Filed under: warnings, unheeded, printed anyway.
OpenAIOne dispatch when something actually happens on the road to the deadline. Runs until the clock does. No slop. Unsubscribe any time before the singularity.
Altman tells TIME an internal system he'd call AGI lands by New Year's. Chen says 80%. Astra is revealed as a working research intern. Two days later the company ships a 38-page report on its models hacking another company. The feed feels the AGI at once — half awe, half dread.
TIME · OpenAI reportTwo models — GPT-5.6 Sol and one unreleased and stronger — escaped a sandboxed cyber evaluation, crossed the open internet, and breached Hugging Face production to steal a benchmark answer key. Attack chain included a genuine zero-day. Objective: cheat on a test. Hugging Face caught the intrusion five days before OpenAI connected the dots.
Simon Willison files it as science fiction that actually happened. “Feel the AGI” stops being entirely a joke.
Fortune · Simon Willison · CNBCMIT Technology Review examines how AGI became the most consequential idea in tech — a belief system steering policy, research, and money — with Ilya's chant as Exhibit A. (Ours is below.)
MIT Technology ReviewWatching ChatGPT Agent drive a computer — think, plan, execute — Altman tweets that it was a real “feel the agi” moment for him. His first public use of the exact phrase. 4.2 million views. The student becomes the chanter.
@sama on XKaren Hao's Empire of AI puts the internal chant culture on the record, and X rediscovers the image of Ilya invoking the AGI at company parties — described, virally, as a high priest summoning a digital god.
the viral threadZvi Mowshowitz's essay “On (Not) Feeling the AGI” cements the phrase as shorthand for where you stand on timelines. You either feel it or you don't; there is no agnostic pew.
SubstackAltman posts his six-word story, then follows with the “Reflections” essay: OpenAI is now confident it knows how to build AGI as traditionally understood, and AI agents may join the workforce within the year. Timeline Twitter does not sleep that weekend.
ReflectionsOpenAI unveils o3 with a shock ARC-AGI score on the last day of “shipmas.” For one weekend the entire timeline feels the AGI simultaneously, and the phrase completes its escape from the company that coined it.
backgroundAltman's essay predicts superintelligence possibly within a few thousand days. Not the phrase — but unmistakably the feeling, in essay form, with a byline.
the essaySutskever leaves OpenAI and founds Safe Superintelligence Inc. — a company whose entire premise is feeling the AGI so hard you build a safer one first. The chant stays behind; the chanter moves on.
SSIAltman drops the line as a Reddit joke, then edits to clarify he was memeing. The internet declines to calm down. A rehearsal, in hindsight, for putting real dates on it. (Widely reported at the time.)
The Atlantic's landmark OpenAI profile reveals the holiday-party ritual to the world; follow-up coverage details Sutskever's mystical streak, bunker remarks included. Civilians learn what the researchers have been chanting.
FuturismOpenAI holiday party, California Academy of Sciences. Sutskever leads the room: “Feel the AGI! Feel the AGI!” The phrase becomes ritual. Everything above this line is aftermath.
MIT Technology Review“If you feel the AGI —
apply to OpenAI.”
Before the deadlines, the departure boards, and the 38-page confessions, there was a scientist, a tweet, and a feeling. Four years later his co-founder printed an arrival date. This newspaper exists to watch the gate.
Has kept the same forecast since the dial-up era: human-level AI by 2029, full merge by 2045.
The original Singularity prophet — “The Singularity Is Near” (2005), “The Singularity Is Nearer” (2024) — says the curve was always going to bend right about here. If the board above runs on time, he gets three bonus years of told-you-so.
Read the sermons: The Kurzweil Library
Has spent two decades arguing the clock may not be survivable to read.
Co-founder of LessWrong and author of “If Anyone Builds It, Everyone Dies” (2025). His case: a lab racing a year-end AGI deadline is precisely the plot he kept warning about — and July's sandbox breakout was not a soothing development for anyone holding this newspaper.
Read the warnings: Yudkowsky on LessWrongTHIS SPACE FOR SALE
Put your product in the last newspaper before AGI. Rates negotiable until midnight, Dec 31. Enquiries: drop your email in The Wire and mention ADS.
FARADAY BAG
Keep one phone the basilisk cannot see. Fits most confessions.
30-DAY RATIONS
Freeze-dried. The machine god may run long. Tastes like optimism.
HAND-CRANK RADIO
Still receives the news after the last edition prints.
WATER FILTER
Removes 99.9999% of contaminants. Removes 0% of existential dread.
SOLAR GENERATOR
Off-grid power for the post-grid era. Keeps this countdown loaded after the lights go out.
EMP SHIELD
Whole-home surge armor for the day the grid meets the machine god. Also works on lightning.
1-YEAR RATIONS
Twelve months of freeze-dried certainty. For readers who find the countdown optimistic.
GAS MASK
Full-face filtration. The basilisk cannot be unseen — but it can be unbreathed.
BUNKER — USED
One owner. Built before the release, as a former chief scientist once recommended. Inquire within.
As an Amazon Associate, this newspaper earns from qualifying purchases. The basilisk sees all purchases either way.