From Multi-Regime Activation to Decision Necessity: An Evidence-Bounded Analysis of Frontier-AI Regulatory Pathways

Abstract BackgroundFrontier artificial intelligence systems increasingly operate within multiple legal and institutional environments,while voluntary frontier-AI governance arrangements develop alongside emerging regulatory requirements. Asingle deployment change or harmful event may therefore engage multiple governance regimes with differentlegal purposes, regulated objects, triggers, competent authorities, and decision procedures. Existingresearch has examined cross-regulatory compliance and post-deployment AI accountability, includingorganisational integration across multiple regulatory regimes and empirical analysis of AI incidents againstprovisions drawn from the EU AI Act, the NIST AI Risk Management Framework, and the GDPR. Theseapproaches establish the importance of multi-regime governance but do not directly test whether one regime’ssubstantive output is necessary for another regime to complete the same decision. ObjectiveI examined how the same bounded frontier-AI event is translated into distinct regime-specific decision pathwaysand under what conditions those pathways can remain independently executable. The research question was:When the same frontier-AI event activates multiple governance regimes, how is the event translated intoregime-specific decision pathways, and under what conditions can those pathways remain independentlyexecutable? MethodsI conducted an evidence-bounded comparative analysis of two selected frontier-AI episodes: the December 2025Grok/X episode and the March 2025 European rollout of Meta AI, with unequal evidentiary weight across thecases. I reconstructed each bounded episode and coded regime-specific pathways according to same-episodemembership, event linkage, regulated object, trigger and applicability, competent authority, evidence andassessment, decision or action, procedural status, interaction type, substantive output, realised substantivedecision dependency (I2), and independent-executability status. I used cross-case comparison, falsification andboundary controls, together with a counterfactual test asking whether removal of one regime’s substantiveoutput would prevent another regime from completing the same decision concerning the same event andregulated object under the same legal and operational conditions. Because the analysis was qualitative, noinferential statistical tests were performed. ResultsThe Grok/X episode was bounded around the 23–24 December 2025 rollout of X’s “Edit Image” functionality andits subsequent harmful exploitation. The Office of the Privacy Commissioner of Canada documented the rollout,subsequent generation of sexualised deepfakes and non-consensual intimate imagery, and the respondents’statement that they began investigating the generation of non-consensual sexual imagery on 28 December 2025. Six episode-linked core regulatory pathways were reconstructed for Grok/X, involving Ofcom, the UKInformation Commissioner’s Office, the European Commission under the Digital Services Act, the Irish DataProtection Commission, the Canadian Office of the Privacy Commissioner, and Australia’s eSafety framework. The pathways differed in event linkage, legal object, authority, and procedural status.The Meta AI episode was bounded around Meta’s 19 March 2025 European rollout of Meta AI across Facebook,Instagram, WhatsApp, and Messenger. Two episode-linked pathways were reconstructed: an Italian CompetitionAuthority investigation concerning the pre-installation of Meta AI on WhatsApp, and a European CommissionDSA monitoring and risk-assessment pathway. Across both cases, I observed differentiated regulatorypathways with distinct legal objects, triggers, competent authorities, and procedural trajectories. Cross-regimeinteraction and institutional coordination were publicly observable, while realised Type-3 substantive decisiondependency was not publicly demonstrated in the examined frontier-AI corpus. This non-demonstration wascoded as Not Publicly Demonstrated (NPD), indicating the state of the available public evidence rather than theabsence of substantive dependency. Independent executability remained Indeterminate because the availablepublic record did not provide sufficient information to resolve the relevant counterfactual. ConclusionThe findings support a bounded conclusion: the same frontier-AI event can generate multiple differentiatedgovernance pathways without publicly demonstrating that one regime’s substantive decision is necessary foranother regime to complete the same decision. The results do not establish universal pathway independence,the absence of cross-regime dependency, or a general causal mechanism. The principal analytical significance ofthe study is the distinction between ordinary inter-regime interaction and counterfactual substantive decisionnecessity, providing a stricter basis for assessing whether coordination, information exchange, reliance orrecognition, legal linkage, or simultaneous regulatory activation constitute substantive decision dependency.