What Exactly Has to Arrive for Us to Call It AGI?

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This article reexamines artificial general intelligence from first principles by asking a question that precedes capability: when we claim that “AGI has arrived,” who or what is the subject of that claim? We argue that many current AGI claims suffer from subject drift. When a system performs well, the combined capabilities of the model, harness, tools, Human participation, and organizational processes are attributed to the model alone. When failure or real-world harm occurs, responsibility is reassigned to the user, platform, tools, or organization. Clarifying the system boundary is necessary, but not sufficient. A complete AGI also requires a “who” that persists across time, allowing identity, memory, knowledge, judgment, action, history, and consequences to remain attributable to the same intelligent actor. We therefore propose and define the AI Actor as the persistent subject of an AGI system. SoulAuth, our open-source Actor-centered identity and authentication infrastructure written in Rust, represents the first engineering step toward establishing this persistent subject in real systems. On this basis, we define AGI as an intelligent actor that persists across time and that, in an open world without a fully predefined workflow, can autonomously understand, judge, choose, act, collaborate, and evolve while incorporating real-world feedback into its continued formation. A Mind OS provides the cognitive order through which this continuity can be maintained across models, sessions, tools, and applications. The article then addresses governance. We argue that governance inside the model is too deep within an increasingly opaque black box, while governance at the final action boundary is too late. We therefore propose an L2 cognitive layer between the model and real-world action, where cognition, judgment, choice, and action take shape. Internal Structured Alignment embeds identity, permissions, constraints, attribution, audit, and responsibility directly into this formation process, allowing a Governance Loop to operate alongside the Capability Loop. The resulting principle is not that humans should control everything AGI thinks, but that AGI judgment must acquire legitimate authority before producing real-world effects: AGI may judge freely, but it may not act freely. Together, persistent subjecthood, the Capability Loop, and the Governance Loop form our integrated philosophical and engineering framework for defining, building, and governing AGI.