Newest papers

Certified Execution Harness for Agentic AI: Safe and Least-Intrusive Petri-Net Supervisory Control Between Reasoning and Action

Chao Gu, Quanyan Zhu · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31376 · cited by 0 works (OpenAlex) · open-access PDF

Spanish

Este trabajo de paso estudia el problema de la ejecución segura de herramientas para los agentes del modelo de lenguaje grande (LLM) bajo inyección rápida. Usando las redes de Petri, formalizamos la armadura de ejecución entre el LLM y sus herramientas como un problema de control de supervisión, en el que las marcas representan capacidades, aprobaciones y presupuestos. Debido a que una transición incendia solo si las marcas que necesita están presentes, una solicitud inyectada cuyas condiciones previas no se cumplen o ya se gastan se rechaza independientemente de su formulación. Tratando los resultados de la herramienta como incontrolables, aplicamos la seguridad por el supervisor autónomo convencional, y esta seguridad se transfiere al sistema en funcionamiento bajo cuatro supuestos. Además, para capturar a un agente que puede proponer

Machine translation (Meta M2M100); the English original follows.

Abstract

This paper studies the problem of safe tool execution for large language model (LLM) agents under prompt injection. Using Petri nets, we formalize the execution harness between the LLM and its tools as a supervisory control problem, in which marks represent capabilities, approvals and budgets. Since a transition fires only if the marks it needs are present, an injected request whose preconditions are unmet or already spent is refused regardless of its wording. Treating tool outcomes as uncontrollable, we enforce safety by the conventional maximally permissive supervisor, and this safety transfers to the running system under four assumptions. Moreover, to capture an agent that may propose again after a refusal, we introduce the notion of a replanning closed loop, in which each retry costs an inference call. We show that, in this loop, the conventional supervisor can be arbitrarily more intrusive than a stricter safe one. Furthermore, for a given proposer model, the maximally permissive least-intrusive supervisor is characterized as a value-threshold refinement of the conventional one. Under precise conditions, this refinement orders two concurrent actions by at most one monitor place. On an autonomous release agent, we find by synthesis that safe deployment requires a step that the supervisor can refuse after a rollout starts.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Supervisor, Supervisory control, Action (physics), Computer science, Petri net, Control (management)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange