The right "not to be subject to a decision based solely on automated processing" is hardening into a new fundamental right. It is not hardening the same way anywhere. The EU is raising it to de facto fundamental status through a four-layer architecture — GDPR Article 22 joined to Articles 14, 26(11), and 86 of the AI Act. Chile constitutionalized it explicitly through its 2021 amendment (Law No. 21.383). The US built it from state-level sectoral statutes that have already started to fall under the Trump administration's preemption pressure. China took the administrative-regulation route through PIPL Article 24. The four patterns diverge at the surface, and they diverge underneath.
This article is the doctrinal inverse of Thaler. In AI copyright, the chance agreement of personality theory and incentive theory built a global convergence. The right against automated decisions runs the other way: the absence of a shared doctrinal anchor builds the divergence. Only the EU holds an anchor — Charter Article 1 (human dignity) and Article 47 (effective remedy) — and the others hold none. Without an anchor, the doctrine cannot consolidate. And even while this piece was being written, the divergence was reorganizing from the inside — the AI Omnibus delay, the Tribunale di Roma's annulment of the Garante fine, and the effective collapse of the Colorado AI Act all happened between March and May 2026.
What the EU's Four-Layer Architecture Decided
GDPR Article 22 carries the substantive right; AI Act Article 14 the provider-side design duty; Article 26(11) the deployer-side notice duty; Article 86 the ex post right to an explanation. SCHUFA Holding (Scoring), Case C-634/21 (CJEU, First Chamber, 7 December 2023, ECLI:EU:C:2023:957), at paragraphs 43–50, held that the probability value generated by an upstream actor is itself a "decision" within Article 22(1), blocking the nominal-human-review workaround. Paragraph 61 invoked Charter Article 47 implicitly through the logic that a restrictive reading would create a circumvention risk and a lacuna — Article 22 taking its place as a mechanism for effectuating Article 47 is the doctrinal mechanism of de facto fundamental-rights status.
Dun & Bradstreet Austria, Case C-203/22 (CJEU, First Chamber, 27 February 2025, ECLI:EU:C:2025:117), at paragraph 59, closed the five-year Wachter dispute: "Those requirements cannot be satisfied either by the mere communication of a complex mathematical formula, such as an algorithm, or by the detailed description of all the steps in automated decision-making, since none of those would constitute a sufficiently concise and intelligible explanation." It proposed a counterfactual-style explanation — a win, in substance, for the legibility camp. Trade secrets are balanced through in camera disclosure to a supervisory authority or court; full algorithmic disclosure was expressly rejected.
The architecture deliberately avoided two areas. Annex III point 2 (critical infrastructure) is carved out by the body of Article 86. Medical-device AI is a structural omission — routed through Annex I of the Medical Devices Regulation, it never enters Annex III at all. This is the point the Kaminski–Malgieri analysis (SSRN 5194301, 8 March 2025) criticizes.
Global Divergence, Surface and Core
Jurisdiction | Doctrinal form | Anchor | Core weakness |
|---|---|---|---|
EU | statutory → de facto fundamental | Charter Arts 1/7/8/47 | medical-device gap |
Chile | explicit constitutionalization | Const Art 19(1) (2021) | confined to neuro-data |
US | state-level statutory (collapsing) | (none) | Colorado effectively neutralized, Apr–May 2026 |
China | administrative regulation + opt-out | (none) | no fundamental-rights frame |
The same risk that misreads Li v. Liu in the Thaler field as "China recognizes AI copyright" runs through PIPL Article 24. The text is nearly identical to GDPR Article 22, but the doctrinal anchor differs — there is no fundamental-rights anchor in a charter or constitution, and no holding of the SCHUFA kind extending upstream scope. It is a regulatory parallel, not a fundamental-rights parallel.
The US Pattern 3 is a demolition in progress. Following EO 14365 (11 December 2025, 90 Fed. Reg. 58499) and the White House recommendations of 20 March 2026, the Colorado AI Act effectively collapsed across April and May 2026 (DOJ intervention 24 April, federal magistrate stay 27 April, replacement SB 26-189 signed by Polis 14 May). It is the empirical confirmation of the proposition that an anchorless statutory regime can fall in roughly five months, from executive order to replacement statute. The Senate's 99–1 stripping of the ten-year moratorium on 1 July 2025 signals that preemption remains hard at the level of federal legislation — only the DOJ-litigation route is a proven tool.
The Test — Some of It Already Under Way
Of four tests, the second has already begun to resolve. (1) The AI Omnibus (political agreement of 7 May 2026) defers high-risk application to 2027.12 / 2028.8 — substantive content preserved, operational timing pushed back. (2) US federal preemption is realized — Colorado has already fallen. (3) Convention 108+ entry into force (34 of 38 states, Moldova the 34th) could build a vector of partial global convergence. (4) Brazil could form a Latin American cluster by combining LGPD Article 20 with the Chilean model.
The Thaler article's jurisprudential core — doctrine shapes the industry — operates here as a second-order extension. Thaler had global judicial convergence build industrial unity; the right against automated decisions has divergence build market divergence. The same vendor bears different duties in the EU, the US, and China. Doctrinal convergence builds industrial unity; doctrinal divergence builds industrial division.
Unlike Thaler, which fixed its consensus through the silence of a denied certiorari, the right against automated decisions has no such moment of institutional closure. The EU's explicit declaration and the US's explicit refusal collide head-on, and even as this piece was being written, Colorado's collapse, the AI Omnibus delay, and the Tribunale di Roma's annulment of the Garante fine ran in parallel. The divergence is expected, in motion, and open to reshaping at any time. Late May 2026 is only one cross-section of it — there is no guarantee this diagnosis still holds in six months, which is the epistemic character of a frontier field of law.