Toward a Formal, Operationalizable Definition of Artificial General Intelligence (AGI)

Background. The Foundational White Paper of AGI Standard (WP-001, 2026) identified the lack of a shared, operationalizable, and jurisdiction-neutral definition of Artificial General Intelligence (AGI) as the first of five structural gaps obstructing credible international governance. While substantive academic and industry definitions have been proposed — from the Turing test (1950) through Legg and Hutter's universal intelligence (2007), the OpenAI Charter formulation (2018), Chollet's skill-acquisition efficiency framework (2019), and the capability-by-generality matrices of Morris et al. (2024) and Turner (2026) — no single definition has achieved cross-jurisdictional convergence, and none is simultaneously operationalizable, scientifically grounded, and suitable for incorporation into binding international instruments. Objective. This Standards Track White Paper, the deliverable of Standards Track ST-1, advances the WP-001 working definition toward full operationalization. It (i) establishes methodological foundations for defining intelligence as a measurable computational property, (ii) compares the principal existing frameworks and exposes their substantive points of agreement and divergence, (iii) refines the AGI Standard operational definition through the specification of a reference cohort, a domain set, a benchmark suite anchor, and measurable thresholds, (iv) introduces the AGI Standard Levels Framework as an independent synthesis suitable for international standards adoption, and (v) provides a legal-formal specification of the definition for potential incorporation into treaty instruments. Approach. The document proceeds through three methodological steps: grounding the analysis in the established formalism of intelligence as problem-solving and the agent–solver distinction; conducting a structured comparative analysis across six principal frameworks; and synthesizing the resulting elements into a unified definition and a seven-level capability framework integrating performance, generality, and agentic capability as three complementary dimensions. Conclusions. The AGI Standard operational definition specifies (a) a multicultural two-tier reference cohort (Core and Extended), (b) a nine-domain set including Physical Interaction and Embodied Reasoning (D9), (c) a benchmark suite anchor combining ARC-AGI-2, Humanity's Last Exam, GPQA, and agentic evaluation protocols, and (d) measurable thresholds supplemented by anti-gaming provisions. The AGI Standard Levels Framework comprises seven levels from Level 0 (Narrow AI) to Level 6 (Artificial Superintelligence). The definition is constructed to remain robust to architectural advances, jurisdiction-neutral for international adoption, and subject to a documented annual and triennial revision cycle. Document ID: WP-002 | Standards Track: ST-1 | Version 1.0 | May 2026Predecessor: WP-001 (DOI: 10.5281/zenodo.20262140)Successor: WP-003 — ST-2 (DOI: 10.5281/zenodo.20363716)Repository: https://agistandard.orgContact: contact@agistandard.orgLicense: CC BY 4.0Founded by Mohammad Elmihamed, 2026.