#Arendt (1 article)

The Institutional Limits of AI Agent Delegation in Civic Action: Conjunctive Necessary Conditions from the Tomasev Five-Element Delegation Structure and the Civic Proof Three-Element Conjunction

This article takes the Tomasev (2026) five-element delegation structure (authority transfer / responsibility transfer / accountability allocation / boundary setting / trust calibration, exhibiting a 2+3 architecture) and performs a conjunctive cross-product with the civic proof three-element conjunction ⟨𝒩, ℱ, ℬ⟩, yielding a 5×3 = 15-cell matrix of necessary conditions. Of these cells, 9 are conditionally satisfiable, 4 are probabilistically degradable, and 2 are structurally unsatisfiable (RT-ℬ ✗ and AA-ℬ ✗). On this basis, civic action is partitioned into three zones — delegable, conditionally delegable, and structurally non-delegable (θ₁ ≈ 0.2, θ₂ ≈ 0.7) — and a further distinction is drawn between permanently structurally non-delegable acts (determined by the philosophical foundations of ℬ) and contextually structurally non-delegable acts (determined by the joint failure of 𝒩 and ℱ). The hard normative constraint imposed by CRPD Art 12 General Comment No. 1 §26–29 — abolishing substituted decision-making — applies to the conditionally delegable zone; this is a binding normative floor, not a legal basis. The distinguishability of supported from substituted decision-making must be established simultaneously across three layers: the ex-ante deliberation layer, the ex-post reversibility layer, and the decision-trace layer. The EUDI Wallet ARF provides multi-profile rather than multi-tenant delegated key custody. CRPD flows back through ICCPR Art 26 and ICESCR Art 9 as a universal engineering obligation binding on all wallet users. Cross-jurisdictional accountability vacuums are further classified into three types — moral crumple zone, algorithmic opacity, and cross-jurisdictional diffusion — and the Article 14 fifth-category gap (Q10) is disaggregated into Q10a (cryptographic attribution of authority transfer) and Q10b (cross-border multi-party accountability allocation). The FTLA-Agent four-tier governance framework (G_industry / G_state / G_recognition / G_oversight) exhibits asymmetric thickness in 2026; a temporal phasing of 5 / 5–10 / 10–15 years is proposed, with a recommended five-party liability allocation of 25 / 25 / 15 / 25 / 10.