The Epistemic Incompleteness Principle

We state and defend the Epistemic Incompleteness Principle (EIP), consisting of two claims. Claim 1 (Incompleteness of Content): no knower can represent all propositions in its domain. Claim 2 (Recursive Non-Certifiability): no internal certification procedure by which a knower estimates the magnitude of its own incompleteness can itself be certified as adequate from within the knower’s representational resources. Every proposed certification procedure is itself an epistemic object requiring certification, generating a regress that does not terminate within the system. Claim 1 is well-established and independently confirmed by Gödel’s incompleteness theorems and recent formal work in AI self-modeling and recognition-based decision theory. Claim 2 is EIP’s distinctive contribution. It does not assert that internal estimates of incompleteness are worthless — they may be useful — but that no such estimate can be certified as adequate from within the system that generated it. The argument for the central premise of Claim 2 takes the form of an exhaustive dilemma: whatever grounds the warrant for any certification procedure must be either internal to the knower, in which case the regress resumes, or external, in which case it fails the criterion of internal certification. No third alternative exists. This recursive non-certifiability arises from the structure of justification itself rather than from diagonalization, and applies wherever justification applies. We examine independent convergences from Kozłowski (2026) and Vadagam (2026), show that neither reaches Claim 2, formalize the regress structure, and develop two principal applications: the structural non-certifiability of AI self-assessment, and the unavailability of any terminating justificatory chain for classical theism’s omniscience claim.