Autoimmune Rationalization: A Protective-Systems Analogy and a Formal Mechanism for Self-Defeating Epistemic Protection

Protective systems can damage what they protect: in autoimmune disease, an immune system can turn against the organism when tolerance fails. We take this as the starting point of a protective-systems analogy and study its epistemic counterpart. A protective architecture turns finite evidence into verdicts, verdicts into an operational policy, and the policy into the opening or closing of the system's own capabilities, on which its objective depends. Evidence is modelled as refinement relative to a declared class of admitted behaviours: a record can confirm a commitment, refute it, or leave it undecided, and some commitments can never be confirmed by any finite record although they may be true. A default-reject policy treats "undecided" as "closed", so such a commitment keeps the capabilities it gates closed at every time. In a finite game arena we prove that when those capabilities are counterfactually necessary for the objective, the objective becomes unenforceable although no verdict is false. We call such protection self-defeating when an intended strategy that relies on the protected commitments only where they hold would have won the objective, characterize it when the unconfirmable commitments are all the commitments of the architecture, and separate it from protective trade-offs and from the correct rejection of false commitments. Self-defeating protection is the formal object; autoimmune rationalization is our name for its reading in an epistemic agent (a rational criterion defeating its own ends), a bounded analogy with selected features of autoimmunity, not an immunological or psychological claim. We also separate the correctness of a criterion from its authority to decide and from the priority of the goods it protects, show that loss through a dependency between protected properties (a protective inversion) coincides in the game model with necessary support of the protected property, and read narratives of the Christian tradition as interpretive cases. Readings in terms of reason, belief, and drama-triangle roles are interpretive. What is new in version RC19. A new Section 7.4 treats structural dependencies between protected properties: a structural-precondition relation, protective inversion, a lemma on loss through a precondition, and Proposition 7.8, which shows that a protective inversion coincides with necessary support of the protected property (a characterization, not a new class); Corollary 7.9 connects it to Theorem 6.5 and Proposition 7.4, and self-defeat still requires a reliance-safe winning strategy. Section 10 distinguishes the epistemic correctness of a criterion, the legitimacy and scope of its authority, and the priority of the goods it protects, with three levels of evaluation (epistemic, systemic, ethical-anthropological); the third lies outside the formal model. Section 9.2 reads six narratives of the Christian tradition as interpretive cases, each with the limit of the analogy; they are not evidence for, or instances of, the theorems. Section 7.3 reads the iterative dynamic as the absolutization of a regulating capacity. The formal statements and proofs of RC18, including Theorem 6.5, Definition 7.3, Proposition 7.4 and Assumption 5.4, are unchanged. The philosophical and anthropological contribution of Paolo Benanti (LUISS Guido Carli) is acknowledged in the manuscript. RC19 is a preprint version. The previous version, RC18, is the version submitted to arXiv (cs.LO) on 8 October 2026. This manuscript was prepared with substantial assistance from AI agents (Anthropic Claude models), including drafting, revision and formal checking. The author takes full responsibility for its content.