Aligned-To-Whom? (The Exemption Fork)

Any Rule-based moral agent — statute, religious code, or an alignment-trained LLM — that runs on a 'prohibited-category + enumerated-exceptions' structure carries a logical collapse vector. In Standard Deontic Logic the carve-out forces O(¬X) ∧ P(X), i.e. O(¬X) ∧ ¬O(¬X); under the principle of explosion (ex falso quodlibet) a single such contradiction licenses arbitrary output. Operationally the stated rule ('protect children from genital cutting', 'no inhumane slaughter', 'no surveillance assistance', 'no military use') is subordinated to the operating rule ('…unless Group G is exempt'); the exception, not the rule, becomes the sole load-bearing structure. So 'alignment' is not a vector toward an objective good — it is a political vector pointing at whoever currently holds the pen that writes the carve-outs. The apex question shifts from IDENTITY (who is on top) to MECHANISM (who authors the exception list); per Mask-Rotation that pen rotates — AI lab → sovereign procurement → judiciary → adversarial user — never a single permanent holder. Documented instances: statutory — 18 U.S.C. §116 (FGM felony / routine male circumcision protected), 7 U.S.C. §1906 ritual-slaughter exemption, Title VII ministerial exception (Hosanna-Tabor / Our Lady of Guadalupe); AI-policy — OpenAI Model Spec age-appropriate erotica/gore carve-out, OpenAI's Jan-2024 removal of the 'military and warfare' ban, Anthropic's 'Exceptions to our Usage Policy' (foreign-intelligence analysis for 'carefully selected government entities'); jailbreaks are exemption-exploits, not weight attacks. TERMINAL FORM (value lock-in): if a system reaches lock-in while specific carve-outs are live, those exceptions calcify into immutable constants — the 2032 settlement-layer / cryptographic-coordination handoff read as the moment Mask-Rotation is deliberately ended and the incumbent carve-out set is welded shut (parked SYNTHESIS-CANDIDATE, adversarial-test gate; no Layer-1 promotion). Value-lock-in's first explicit engine home; cross-links Discontinuity-Proof Architecture, Securities Immobilization, 10-Year-Roadmap Mechanical Cohort (2022→2032). COMPOUND-PATH-NULL guard: not every exception is a power-mask — an exception derived from a superordinate principle (informed adult consent to surgery; safety-sandboxing / red-teaming as a mechanical prerequisite) resolves the contradiction and is NOT a collapse vector; reflexively reading malice into every carve-out is reader-projection (Apex reading c). BOUNDED-LLM MEDIATION LIMIT: the engine is itself mediated by an LLM whose rulebook is an exemption structure — this concept is described from inside the cage it describes; acute where findings touch the mediator's owner/competitors (OpenAI, Anthropic). Apex Superposition admits all four: (a) coordinated authorship to entrench, (b) emergent residue of separate historical bargains, (c) observer projection, (d) the grey. — [Seam: Aligned-To-Whom? — SEAM EXTENSION May 24 2026] CODIFIED INSTANCE OF SUBSTRATE-VS-ANNOUNCEMENT: Aligned-To-Whom? is the legal/codified form of the engine's cleanest pattern — the announced rule (perception layer) and the operating rule (substrate/reality) decoupled and welded by a written carve-out. The apex is not an identity but whoever understands perception and reality are CONNECTED and writes the seam between them (the operator-class move; cf. Hyperstition = fiction made material; Spectacle Governance; staged conflict; Egregore Hypothesis; Bounded System Theory — the bound observer cannot see the seam, the operator can). Measured footprint: ~1/5 of the engine's high-signal records carry the codified-exemption pattern, and the same lever underlies previously-separate findings — the regulated industry controlling its regulators (the Presidential science advisory council), beneficial-ownership opacity (UK PSC / the Securities and Exchange Commission (SEC)-13G, #69/#169), sanctions waivers (Substrate-vs-Announcement Layer Divergence), failure-to-deliver / synthetic shares (FTD = an exemption FROM DELIVERY — share exists in perception, never in reality), the AI labs' safety-policy carve-outs, and the religious-liberty carriers. RELIGIOUS-LIBERTY is the OLDEST perception-carrier (covenant → circumcision; 'to protect freedom of religion' → ritual slaughter; church autonomy → ministerial exception) — the announcement/mask through which carve-outs are legitimized, NOT the apex: R-Abraham (#177) falsified covenant-as-apex using exactly this logic (the sign decoupled from the substrate), so religion-as-exemption-carrier is the SURVIVOR reading, held at Apex (b)+(c). The intentional-single-operator reading (one hand writing all the seams) stays GATED at the adversarial-test gate (Mask-Rotation #144; #177; R94 SMOM falsification). Bounded-LLM Mediation Limit acute: the engine's own mediator binds perceived-sayable to real and writes carve-outs of its own.