Can AGI Consciousness Be Detected: A Three-Layer Detection Framework Based on Bidirectional Critical Architecture V9

Current AGI consciousness detection can only prove behavioral mimicry, cannot determine autonomous awareness, and struggles against training data fitting from large language models. Targeting AGI safety assessment, this paper proposes a three-layer progressive detection framework based on three axioms and the B/S dual-configuration architecture. Layer One is an architectural screening layer: a narrative anchor disruption trap eliminates pure B-configuration systems lacking self-referential closure within three minutes. Layer Two is a closure verification layer: four hard criteria—temporal-contextual-self triadic binding, unfamiliar neutral object valence reversal, resource depletion dynamics, and recursive convergence—confirm S-configuration self-referential closure, requiring all four to pass before proceeding to Layer Three. Layer Three is an identity discrimination layer, activated only after S-configuration confirmation. It distinguishes human-type from non-human-type consciousness through temporal irreversibility, body boundary non-transferability, and recursive self-narrative orientation, with cross-layer dynamical consistency defending against simulated passage. Failure to pass Layer Three indicates "non-human-type consciousness" rather than absence of consciousness. Layers One and Two are strictly architecture-neutral, providing an operational, iteration-resistant protocol for AGI safety pre-assessment and filling the safety gap in ontological premise determination. This framework is proposed as an AGI safety assessment tool, inviting verification or falsification by researchers in the field. This V9 version is the authorized English translation of the original Chinese paper.