Toward a Unified Global Framework for Artificial General Intelligence Standards (AGI)
Year: 2026
Type: article
Abstract: Background. Artificial General Intelligence (AGI) is widely regarded as among the most consequential technological developments of the twenty-first century. Despite accelerating capability gains across frontier AI systems, no internationally agreed definition of AGI exists, no standardized evaluation protocols have been ratified, and no binding multilateral instrument governs its development or deployment. A growing body of expert opinion characterizes this governance vacuum as a source of significant societal-scale risk warranting urgent international attention; this view is, however, contested in the scholarly literature, and the present document engages with that disagreement explicitly. Objective. This Foundational White Paper establishes the institutional, conceptual, and methodological basis for AGI Standard — an independent open global research initiative dedicated to developing unified and safe standards for AGI. The document articulates the rationale for unified standards, identifies five structural gaps in the current AI governance landscape, and proposes a layered standards architecture composed of definitional, evaluation, governance, and equity components. Approach. AGI Standard organizes its work around six interlocking pillars — Unified Standards, Humanity Protection, Global Collaboration, Radical Transparency, Equity and Justice, and Responsible Innovation — operationalized through five Standards Tracks (ST-1 to ST-5) addressing the formal definition of AGI, evaluation protocols, international legal architecture, equity and distribution, and catastrophic risk taxonomy. Each Standards Track is mandated to produce a successor white paper (WP-002 through WP-006). Conclusions. Voluntary commitments by frontier AI developers, while constructive, cannot substitute for binding international standards. The institutional architecture required to govern AGI must be established before AGI-class capabilities are achieved. AGI Standard invites the global scientific, legal, and policy community to participate in the open development of this architecture, ensuring that AGI — when it arrives — serves the whole of humanity rather than a narrow subset of it. Key Contributions 1. A preliminary working definition of AGI, characterized by the autonomous acquisition, integration, and application of knowledge and skills across an open-ended range of cognitive domains — including the autonomous pursuit of long-horizon goals — at or above the level of a competent adult human professional, measured against a defined reference cohort and a published benchmark suite, without task-specific architectural modification or weight retraining. The definition is positioned against five principal existing frameworks, from Turing (1950) to Morris et al. (2024). 2. A diagnosis of five structural gaps in current AI governance — insufficient convergence on a universal AGI definition; insufficient international evaluation standards; the absence of a binding international legal framework; insufficient inclusive representation and equitable distribution; and insufficient structured anticipation of catastrophic risk — each addressed one-to-one by a dedicated Standards Track. 3. A transparent standards-adoption procedure: draft standards are open for public comment for a minimum of ninety days (thirty days for non-substantive revisions), every substantive comment receives a published response, and adoption proceeds by consensus. Independence. AGI Standard does not represent any government agency, commercial company, or established international organization, and this work received no external funding. Document ID: WP-001 | Foundational White Paper | Version 1.0 | October 2026Author: Mohammad Elmihamed — Founder, AGI StandardWebsite: https://agistandard.orgContact: [email protected]: CC BY 4.0
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7223456171
FWCI:
Cites:
Cited by:
Related to: