Not a Divisible Good: Erman and Furendal's Legitimacy Chain Beyond Its Own Scope - Testing an AI-Governance Framework Against Five Structural Paradoxes of AI-Driven Longevity Biopolitics Allocation

Eva Erman and Markus Furendal's "Artificial Intelligence and the Political Legitimacy of Global Governance" (Political Studies, 2022/2024) is, on this corpus's own assessment, the most precise account currently available of what makes a governance body legitimate rather than merely effective, and this paper's argument depends on that assessment being correct rather than being softened for the sake of contrast. Their apparatus distinguishes governance of AI, the regulation of AI's own development and deployment, from governance by AI, the use of AI within existing governance structures; it distinguishes authorized entities, whose right to make coercive law derives from democratic procedure, from mandated entities, which receive delegated, non-coercive policymaking authority and remain legitimate only so long as a genuine right of revocation is retained by whoever delegated it; and it requires the resulting legitimacy chain to be secured by transparency, so that revocation can be meaningfully exercised, and publicity, a right to justification for individual decisions. Applied to the landscape of actually existing global AI governance — the GDPR, the EU AI Act, NATO's AI strategy, the soft-law ethics guidelines issued by AI labs and industry bodies such as IEEE, IBM, DeepMind, and the Future of Life Institute — this apparatus produces a genuinely important negative finding: non-state private actors issuing guidelines that nobody delegated to them and nobody can revoke constitute a legitimacy deficit, whatever the quality of the guidelines' actual content. This paper does not dispute a clause of that finding. What Erman and Furendal's 2022/2024 paper does not do, because the case in front of them never required it, is examine AI-driven longevity governance specifically: a domain their framework's stated scope covers in principle, since it is a case of AI governance, but whose specific dynamics none of their published case studies touch, because none of the GDPR, the AI Act, NATO's strategy, or algorithmic administrative decision-making involves allocating a scarce, life-determining biological intervention under the conditions this corpus's own Biological Zero-Day Mechanism formalizes: discreteness, permanence, and institutional authorship. This paper's question is offered as an extension of their project, not a rebuttal to it: does the legitimacy chain built for AI governance's examined cases transfer cleanly to this further case, or does AI-driven longevity allocation belong to a different category of political good, one their apparatus, reasonably, was never asked to handle? The paper's diagnostic distinction, developed in Section I and independently confirmed by Guido Calabresi and Philip Bobbitt's 1978 theory of tragic choices, is that democratic-legitimacy theory generally, including E&F's own apparatus, is built for goods that are divisible (allocable at the margin), iterable (this cycle's loser contests the next), and recoverable (a wrong decision can be reversed). The Biological Zero-Day Mechanism's own formal properties make longevity-relevant allocation none of these things in the timeframe that determines who lives. The paper then runs each of this corpus's five structural paradoxes against a distinct, specific component of E&F's apparatus, to locate precisely where and how the transfer breaks down rather than asserting a general mismatch. First, the Governance Paradox is run against the democratic minimum: it shows that equal opportunity to participate does not produce equal access to a capacity-constrained slot, and that the specific reason this fails is structural rather than contingent — A5 is not a preference that deliberative processes were built to transform, because it is constitutive of the participants themselves in the sense Bernard Williams identifies for ground projects, and deliberative iteration presupposes precisely the kind of preference-transformability under public reason that constitutive attachments do not have. Second, the Reform Acceleration Paradox is run against E&F's implicit assumption that increasing transparency, revocability, and publicity monotonically increases legitimacy and public acceptance: it shows this assumption, exactly right for the divisible goods E&F examine, can reverse under zero-sum stakes, because greater transparency converts diffuse, unattributed exclusion into specific, attributable decision, and attributable decision is what activates A5's political sequence. Third, the Generative Paradox is run against E&F's own implicit recommendation that a body like the AI for Good Global Commission be strengthened into a genuine, accountable, mandated authority: it shows that Michels's iron law of oligarchy operates with maximal force specifically in this domain because longevity allocation uniquely combines four factors — extraordinary resource value, extreme technical specialization, explicit zero-sum structure, and permanent stakes — that Michels's mechanism requires to be jointly present to produce durable capture, and that E&F's apparatus has no built-in resource for detecting this conjunction. Fourth, the Dam Paradox is run against the revocability condition: it shows revocability has always depended on an unstated second mechanism, biological turnover among officeholders and elites, that BZM disables selectively for whichever class is best positioned to occupy legitimate seats over multi-decade horizons. Fifth, the ARSI–Longevity Resonance Paradox is run against the deliberative process itself: it shows that under λ_bz's documented compounding, a fully legitimate authorization-delegation-revocability cycle takes longer to complete than the frontier it is adjudicating remains current, so that legitimate allocation of frontier F(t) arrives only after the operative frontier has already become F(t+Δ). Two further findings, developed in Sections I, IV, and VIII, confirm this extension rather than merely asserting it, and both are offered in a spirit of continuity with Erman and Furendal's own project rather than correction of it. Calabresi and Bobbitt's independent 1978 study of tragic allocation — draft conscription, dialysis machines, childbearing limits — found that societies facing indivisible, irreversible allocation problems do not converge on one stable, fully legitimated method and hold it, but cycle between methods, because sustained visibility of the same method's cumulative toll is what societies cannot bear, a finding arrived at from law and economics, decades before BZM existed, that independently locates the same category boundary this paper is describing. And Erman's own 2020 foundation, which holds that legitimacy principles are function-sensitive rather than uniform across every type of decision, already contains the resource for this paper's extension: if lawmaking, policy-making, implementation, and monitoring each properly receive their own principles, then whether terminal, one-shot allocation of a life-determining good constitutes a further function-category, requiring principles of its own, is a question her own method invites. This paper does not derive those further principles. It shows, using Erman's own method and reaching the same boundary Calabresi and Bobbitt reached from outside political theory entirely, that they are the next, natural extension of a framework this paper treats throughout as correct, necessary, and worth building on rather than replacing. Two developments since this project's initial drafting confirm the argument empirically rather than only structurally, and Section V discusses both in full. The UN and International Telecommunication Union's AI for Good Global Commission — launched 1–2 July 2026, holding its inaugural session 8 July 2026 in Geneva, composed of heads of state alongside the chief executives of Nvidia, Amazon, Microsoft, Anthropic, and Cohere — is the exact strengthened, high-composition-legitimacy body Section V anticipates, and its own inaugural output, a call for action rather than binding commitments with named parties and timelines, resolved in the direction the Generative Paradox predicts. Dubai's Law No. 17 of 2026, issued 10 June 2026, establishing the Dubai Longevity Authority, is the first governance body formed anywhere specifically to regulate longevity access — and its founding statute names international investment attraction as a formal objective coordinate with, not subordinate to, public health. Neither development is evidence of bad faith by any named actor. Both are the paper's theoretical prediction arriving before the paper itself was finished.