Auditable AGI Claims: Generality, Actor Boundaries, and Evidence Obligations for the Public Reference Layer

When technical statements such as “AGI has arrived” or “this system is approaching AGI” are reused as factual premises in subsequent research, media coverage, market judgments, policy reasoning, or governance discussions, conceptual ambiguity, boundary drift, and evidential extrapolation cease to be merely terminological problems. They become problems of public information infrastructure. We call the context in which such technical conclusions are subsequently reused by other social actors as factual premises the public reference layer. This paper asks what information an AGI claim must disclose about its semantics, evaluated object, scope, evidence, and update conditions in order to become a research object that third parties can compare, challenge, revise, and cite with appropriate caution. We do not propose a unique ontological definition of AGI, nor do we offer an ultimate adjudication of its truth. Instead, we introduce a registration and audit protocol for the public reference layer. The protocol treats a versioned, specific claim as the unit of analysis and distinguishes four meanings of generality: cross-task diversity (G1), cross-distribution transfer (G2), task-frame generation (G3), and sustained adjudication under normative gaps (G4). It also requires explicit object boundaries, evaluation scope, success and failure criteria, the roles of Humans and external mechanisms, and the evidence-access conditions actually available. These four meanings distinguish different commitments; they do not constitute a demonstrated capability hierarchy. Evidence for one meaning of generality cannot automatically substitute for evidence required by another, and the scope of a conclusion must not exceed the scope actually supported by the evidence. For G4, we unpack internal adjudication, diachronic continuity, and the actual participation of the stated adjudicative grounds into inspectable structural commitments and evidentiary obligations, while separating facts about the object from the materials available to the auditor. For every subclaim that proceeds to substantive evaluation, the protocol returns one of three statuses: Supported, Refuted, or Unverified. Versioned records further constrain attempts to preserve an earlier conclusion by retrospectively shifting definitions, objects, scope, or success criteria. A fully specified constructed case demonstrates the protocol’s applicability and shows that its outputs can be recomputed from fully specified inputs. We also provide a reusable DCT registration form, audit card, status-update rules, and a normative-gap test template. The paper proposes and formalizes a public audit protocol that is adoptable, testable, falsifiable, and revisable. It does not certify any real-world system as AGI, nor does it claim that the protocol’s real-world deployment effects or inter-auditor agreement have already been empirically validated.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-12
DOI
https://doi.org/10.5281/zenodo.24163986
Primary Topic
Ethics and Social Impacts of AI
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Auditable AGI Claims: Generality, Actor Boundaries, and Evidence Obligations for the Public Reference Layer

Lucas Luo, Yuan Kun, Kiki Hu, Egusi Gui et al.
Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
preprint

Auditable AGI Claims: Generality, Actor Boundaries, and Evidence Obligations for the Public Reference Layer

Lucas Luo, Yuan Kun, Kiki Hu, Egusi Gui, Magnus Hu, Harold Wang, Echo Li
preprint en

Abstract

When technical statements such as “AGI has arrived” or “this system is approaching AGI” are reused as factual premises in subsequent research, media coverage, market judgments, policy reasoning, or governance discussions, conceptual ambiguity, boundary drift, and evidential extrapolation cease to be merely terminological problems. They become problems of public information infrastructure. We call the context in which such technical conclusions are subsequently reused by other social actors as factual premises the public reference layer. This paper asks what information an AGI claim must disclose about its semantics, evaluated object, scope, evidence, and update conditions in order to become a research object that third parties can compare, challenge, revise, and cite with appropriate caution. We do not propose a unique ontological definition of AGI, nor do we offer an ultimate adjudication of its truth. Instead, we introduce a registration and audit protocol for the public reference layer. The protocol treats a versioned, specific claim as the unit of analysis and distinguishes four meanings of generality: cross-task diversity (G1), cross-distribution transfer (G2), task-frame generation (G3), and sustained adjudication under normative gaps (G4). It also requires explicit object boundaries, evaluation scope, success and failure criteria, the roles of Humans and external mechanisms, and the evidence-access conditions actually available. These four meanings distinguish different commitments; they do not constitute a demonstrated capability hierarchy. Evidence for one meaning of generality cannot automatically substitute for evidence required by another, and the scope of a conclusion must not exceed the scope actually supported by the evidence. For G4, we unpack internal adjudication, diachronic continuity, and the actual participation of the stated adjudicative grounds into inspectable structural commitments and evidentiary obligations, while separating facts about the object from the materials available to the auditor. For every subclaim that proceeds to substantive evaluation, the protocol returns one of three statuses: Supported, Refuted, or Unverified. Versioned records further constrain attempts to preserve an earlier conclusion by retrospectively shifting definitions, objects, scope, or success criteria. A fully specified constructed case demonstrates the protocol’s applicability and shows that its outputs can be recomputed from fully specified inputs. We also provide a reusable DCT registration form, audit card, status-update rules, and a normative-gap test template. The paper proposes and formalizes a public audit protocol that is adoptable, testable, falsifiable, and revisable. It does not certify any real-world system as AGI, nor does it claim that the protocol’s real-world deployment effects or inter-auditor agreement have already been empirically validated.

Zenodo (CERN European Organization for Nuclear Research)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon