Abstract

Contemporary approaches to Artificial General Intelligence (AGI) rely predominantly on behavioral benchmarks, optimization performance, and scaling laws. This paper argues that such approaches are ontologically insufficient. Intelligence cannot be verified through behavior or task performance; it can only be falsified through the presence of forbidden ontological structures. We introduce the Ontological Test Suite (OTS), a set of falsification-based criteria designed to identify systems that are ontologically incapable of intelligence, regardless of their performance, creativity, or utility. Grounded in a minimal process ontology (Metamonism CORE v1.3), we formalize intelligence as the continuous impossibility of final stabilization and introduce Unfold as a mandatory, non-representable rupture preventing cognitive selfidentity. OTS does not confirm AGI; it excludes non-AGI. This reframes the AGI problem as one of ontological viability rather than architectural ingenuity.