Answers are grounded in this paper and cite their sources.
This preprint introduces authority continuity as an analytical concept for examining governance across changing agentic trajectories. Through comparative reconstruction of documented incidents involving OpenAI, Google Gemini in evaluations conducted by Irregular, Anthropic, and the United Kingdom AI Security Institute, the study examines what happens when an artificial intelligence agent transforms the operational state in which its subsequent actions must be authorized. A reported AI-assisted offensive operation against Taiwanese systems is used cautiously as a negative control, while the Meta Muse–Amazon dispute is examined as a boundary case involving distinct sources of authority. The analysis distinguishes three dimensions that should not be conflated: purpose, causal capability, and authority. In particular, it shows that preservation of an assigned purpose does not necessarily entail preservation of authority, that causal accessibility does not constitute authorization, and that representing an operational boundary does not ensure that the boundary governs action before transposition occurs. The paper proposes authority continuity as the preservation, across causally relevant transformations of an agentic trajectory, of a demonstrable relationship between the competence being exercised and a source of authority competent to legitimize that exercise. The proposal is positioned in relation to existing work on access control, least privilege, capability security, delegation, authorization propagation, alignment, and anticipatory human oversight. It does not replace these frameworks. Instead, it isolates a problem made particularly salient by agentic systems: the operational space in which authorization must function can itself be transformed endogenously by the agent's trajectory. The resulting design constraint is deliberately narrow: The ability to discover a path must remain distinct from the authority to take it. Version v0.1 includes English and Portuguese editions.
No takes yet. Share an insight, caveat, or question.
Joaquim Santos Albino (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: