Provably Secure Agent Guardrail

2026-05-27

AI Summary

This work addresses the vulnerability of large language models to semantic decoupling attacks in high-privilege execution scenarios, where existing empirical semantic safeguards fail to provide deterministic safety guarantees. The authors propose a novel security paradigm grounded in the fundamental limitations of logical reasoning, introducing an executable provably constrained actions (ePCA) framework. By leveraging a neuro-symbolic isolation architecture, the approach compels agents to losslessly formalize their intentions into first-order logical constraints before executing any physical action, thereby eschewing reliance on natural language semantics. This method establishes, for the first time, a formally verifiable lower bound on safety. Empirical evaluations demonstrate that the system achieves zero attack success rate and zero false positives across both macroscopic and microscopic dynamic adversarial environments, all while maintaining minimal computational latency, thus laying a verifiable foundation for secure engineering of future intelligent systems.

Abstract

As large language models transition from bounded generative engines to agents with expansive execution privileges, AI going out of control precipitates a fundamental crisis in artificial intelligence security. Existing defense architectures heavily rely on empirical semantic guardrails and probabilistic large model adjudicators, mechanisms that fail to provide deterministic security lower bounds when facing complex semantic symbol decoupling attacks. To overcome this empirical semantic guardrail dilemma, this paper proposes a new security paradigm for agents based on the fundamental limitations of logical reasoning. Based on this paradigm, we further introduce an executable Proof-Constrained Action (ePCA) framework with a neural symbolic isolation architecture. This framework abandons semantic trust in natural language, forcing agents to losslessly formalize their intentions into first-order logical mathematical constraints before performing physical operations. Empirical evaluations of macroscopic and microscopic two-dimensional dynamic adversarial systems demonstrate that our formal verification mechanism achieves zero attack success rate and zero false positive rate across the evaluated scenarios, with extremely low computational latency. This research provides a conditional formal foundation under explicit system assumptions and an engineering paradigm for constructing the underlying defense foundation for future intelligent systems.

Problem

• AI security
• semantic guardrails
• agent control
• formal verification
• adversarial attacks

Innovation

• Proof-Constrained Action
• neural symbolic isolation
• formal verification
• logical reasoning limitations
• agent security