When Instructions Stop Governing: Rethinking Control in AI Agents

Timothy M. Rogers

PAPER · v1.0 · 2026-09-17 · human

Formal Sciences Computer Science Artificial intelligence and machine learning

Abstract

Recent failures of AI-agent control become easier to understand—and potentially easier to design against—once control is reframed not as the presence of instructions, but as the preservation of which constraints actually govern an agent’s evolving activity.

Keywords

Hierarchical Relational Ontology AI-agent control Governing Constraints Semiotic logic

Download PDF