AGI timeline: every public forecast, dated, with the skeptics

The AGI timeline, meaning the dates people have publicly given for artificial general intelligence, is not one forecast but a pile of them, made against different definitions. This page lists the notable ones with dates, who made them and what they actually said, then notes where the skeptics stand. It is a tracker: we add new forecasts and revise old entries when their authors do. Last updated October 1, 2026.

First, a rule for reading it. A date means little without a definition. Our glossary, AGI vs ASI, explains why. Where a forecaster uses a narrower milestone, such as a coding system, we say so.

The researchers' baseline: 2022 and 2023 surveys

AI Impacts surveyed 2,778 authors who had published at top AI venues in October 2023. The aggregate forecast gave a 10% chance that unaided machines outperform humans at every task by 2027 and 50% by 2047, which was 13 years earlier than the equivalent 2022 survey (2060). The same survey gave 10% odds that all human jobs are fully automatable by 2037, and 50% by 2116, against 2164 in the 2022 survey. It is a researcher survey, not a lab forecast, and it is almost three years old.

The lab leaders

• Sam Altman, OpenAI, June 2025. In "The Gentle Singularity" he wrote: "We are past the event horizon; the takeoff has started." On dates: "2026 will likely see the arrival of systems that can figure out novel insights" and "2027 may see the arrival of robots that can do tasks in the real world." He wrote that "in the 2030s, intelligence and energy" will become "wildly abundant." He gives no single AGI date, so we log it as a direction.
• Mark Zuckerberg, Meta, July 30, 2025. "Developing superintelligence is now in sight." A statement of position, with no date.
• Dario Amodei, Anthropic, January 2026. In "The Adolescence of Technology" he defines powerful AI as a model "smarter than a Nobel Prize winner across most relevant fields" and says it could be as little as one to two years away, which points to about 2027 or 2028 from his writing date. He stresses uncertainty.
• OpenAI leadership, 2023. Wikipedia reports that they said superintelligence "may happen in less than 10 years." We have not traced that to the original post, so treat it as secondary.

The forecasters

AI 2027, April 3, 2025. Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean published a scenario with milestones: superhuman coder in March 2027, superhuman AI researcher in August 2027, superintelligent AI researcher in November 2027 and artificial superintelligence in December 2027. A note dated November 22, 2025 says "2027 was our modal (most likely) year at the time of publication, our medians were somewhat longer," and points to a model update that pushed the median back about 1.5 years while keeping a 2027 superhuman coder "a serious possibility."

AI Futures Project, August 16, 2026. Eli Lifland, Daniel Kokotajlo and Brendan Halstead gave median (p50) dates for an automated coder, a system leading companies would prefer over human programmers: Kokotajlo November 2027, Halstead January 2029, Lifland January 2030. They write that reality is progressing at about 70% to 90% of the pace in their earlier scenario. This is a coding milestone, not full AGI.

The 2026 shift. The 80,000 Hours podcast "What the hell happened with AGI timelines in 2026?" reports that people who updated between January and April 2026 moved sooner, citing among other things faster METR time-horizon doubling and revenue growth. It records Rob Wiblin moving from shocked at fully automated AI research in 2027 (January 2026) to calling it imaginable in 2027 and plausible by 2028 (July 2026). These are commentators' judgments on evidence we have not rechecked here, and the show lists factors that still support longer timelines into the mid-2030s. FutureSearch's tracker shows most forecasters moved sooner from 2023 to 2025, with exceptions such as Tamay Besiroglu, and then a mixed pattern in 2025 and 2026.

The skeptics

Yann LeCun, January 23, 2024, Davos. Reaching human-level AI "will take years, if not decades" and "it's not around the corner," requiring "new scientific breakthroughs that we don't know of yet." He listed what current systems cannot do: understand the physical world, remember, reason and plan. That was over two and a half years ago, and we have not located a dated 2026 LeCun quote we could open, so it is logged as his last verified one.

Where the dates cluster

• 2027 or 2028: Amodei (powerful AI), Kokotajlo (automated coder, November 2027).
• 2029 to 2030: Halstead and Lifland (automated coder, January 2029 and January 2030).
• 2047: the 2023 researcher survey, at 50%.
• Decades: LeCun, 2024.

What to watch

We will add a row whenever a lab chief, a forecasting group or a survey publishes a dated number, and we will mark entries superseded when authors revise. The next scheduled item is not a forecast but a policy date: the science adviser's proposed definition of "Super Intelligence" is due around November 28, 2026 under the executive order, and a legal definition would give timelines something fixed to point at. For the word itself, see what is superintelligence.

Our take

The honest range runs from 2027 to "decades," and the spread is mostly definitions. Boosters have moved dates sooner over the past year while some forecasters have kept hedges in their own text. We would trust a forecast more when it names a measurable milestone, as the AI Futures Project does with an automated coder, and less when it says "soon." Treat every row as a claim by an interested party until a test says otherwise.