AGI would arrive by 2030
Under review The claim cannot be scored yet and may never be, because no settlement definition exists. The trackability analysis is the deliverable: every prior decade-scale Altman timeline is still open, and the nearest measurable proxies, benchmark suites and economic task automation, remain unsettled.
Show 5 supporting points
- No agreed definition: AGI, superintelligence, and "extraordinarily capable" are used interchangeably across Altman's public statements, so no binary outcome is defined.
- February 2023: "Planning for AGI and beyond" describes preparations for AGI without a date, which keeps the 2025 claim from being compared against a prior miss.
- 2024 MIT interview: Altman said the term AGI has been "overused", a definitional concession that makes the 2030 framing hard to pin down.
- September 2024: "The Intelligence Age" says "it is possible that we will have superintelligence in a few thousand days (!)", roughly 2030-2035 counting from publication, a window that overlaps but does not equal 2030.
- Nearest testable proxies, such as AI completing a defined share of economic tasks, have no published settlement source as of the retrieval date.
I would certainly say by the end of this decade, so, by 2030, if we don't have models that are extraordinarily capable and do things that we ourselves cannot do, I'd be very surprised.
Altman, September 2025, Axel Springer Award interview in Berlin. The claim's testability problem is definitional: "extraordinarily capable" has no agreed operational meaning, and AGI has shifted in his own usage from a term he once called overused to a near-term expectation.
What happened, dated
Every entry carries the date and the record behind it. Sources are indexed against the register below.
-
Planning for AGI
Altman publishes a plan for AGI development and safety with no date attached. Source 2
-
Term questioned
In interviews around 2024, Altman says AGI has been overused as a term and resists defining a threshold. Source 3
-
A few thousand days
"The Intelligence Age" describes superintelligence as possibly a few thousand days away. Source 4
-
By 2030
Axel Springer interview: he would be very surprised if models are not extraordinarily capable by the end of the decade. Source 1
-
Window closes
The claim becomes scoreable only if a definition is fixed before the deadline. Source 1
What the evidence shows
The moving definition is the problem
Score the 2025 statement against its own words and the definition moves. "Extraordinarily capable" can be satisfied by any impressive model, which is exactly how the term AGI drifted from overused to imminent in two years. A claim that redefines itself is not trackable until the definition is frozen.
A few thousand days is not 2030
The September 2024 essay points somewhere in the early-to-mid 2030s, the September 2025 interview at 2030. These are different predictions under the same brand. The audit records both because the difference matters: a reader betting a deadline on "few thousand days" and one betting on 2030 are taking different positions.
How JMM would settle it
Three frozen pieces would make this scoreable: a published definition, a test suite, and a named evaluator. Until OpenAI publishes that contract, the honest label is under review, and the useful output is the map of what each version of the claim would require to be proven.
Every receipt, with retrieval dates
JMM attaches a statement only when the contemporaneous primary record and an exact locator exist. Retrieved dates are when JMM last verified each link.
Evidence strength: Quoted from the primary record. Qualified: no official or statistical series is attached, so the verdict rests on the documentary record alone.
Axel Springer interview quotes including the by-2030 statement
Opening paragraphs on AGI preparation without dates
Coverage of the by-2030 quote and its context
"It is possible that we will have superintelligence in a few thousand days (!); it may take longer …"