TLDR
Artificial general intelligence (AGI) is a hypothetical AI that can learn and perform any intellectual task a human can, across domains, rather than one narrow task. There is no agreed definition or test, and whether current systems count is disputed (see below).
Definitions in use
- Capability levels (Google DeepMind, 2023): “Levels of AGI” grades systems by performance (emerging, competent, expert, virtuoso, superhuman) and generality; it classed ChatGPT and LLaMA 2 as “emerging AGI” (per Wikipedia’s summary).
- Contractual definition: the 2025-10-28 Microsoft-OpenAI agreement says that once OpenAI declares AGI, the declaration “will now be verified by an independent expert panel”; Microsoft’s research IP rights run until the panel verifies AGI or through 2030, whichever comes first. AGI is therefore also a commercial trigger, not only a research concept.
- Benchmarks: ARC-AGI (ARC Prize) measures skill-acquisition efficiency on unseen tasks, “easy for humans, hard for AI”; per arcprize.org (fetched 2026-10-05) it has ARC-AGI-1, -2 and -3, with ARC Prize 2026 running. Benchmark success is not by itself AGI.
Where the debate stands (2026)
- Labs disagree: Wikipedia reports Sam Altman saying in December 2025 that “we built AGIs” and that AGI “went whooshing by” with less societal impact than expected (secondary source).
- The International AI Safety Report 2026 (2026-02-03, chair Yoshua Bengio) describes uneven capabilities: models pass professional exams and reach IMO gold-level maths yet fail simple tasks, and it notes capability gains now come increasingly from post-training techniques, not scale alone. Stanford’s 2026 AI Index (2026-04-13) is reported to make the same “jagged frontier” point (not rechecked).
- Forecasts are survey-based and scattered; per Wikipedia (secondary), older polls (2012-2013) had a median for 50% chance of AGI around 2040-2050, and a September 2025 analysis puts typical expert views near 2040. Treat all dates as opinion.
Why it matters
In the author’s view, AGI claims influence investment, contracts, safety policy and regulation: see ai-safety-and-governance (frontier lab safety frameworks), ai-regulation-and-policy, ethics-in-ai, future-trends-in-ai and the labs in ai-research-labs. Related concepts: artificial-intelligence, large-language-model.
Open items
- No consensus test; Altman/Brockman-era claims (including a “GPT-6” AGI-era statement on Wikipedia) were not checked against OpenAI primary sources and are not repeated as fact.
Sources (fetched 2026-10-05)
- https://en.wikipedia.org/wiki/Artificial_general_intelligence (secondary)
- https://blogs.microsoft.com/blog/2025/10/28/the-next-chapter-of-the-microsoft-openai-partnership/
- https://arcprize.org/arc-agi
- https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026
- https://hai.stanford.edu/ai-index/2026-ai-index-report