arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.22136cs.CRcs.LG
收藏

主权代理循环:在现实系统中解耦AI推理与执行

Sovereign Agentic Loops: Decoupling AI Reasoning from Execution in Real-World Systems

Jun He, Deying Yu

AI总结:

本文提出Sovereign Agentic Loops,通过解耦AI推理与执行,提升系统安全性,实验显示其有效阻止93%的危险意图。

摘要:

大型语言模型(LLM)代理越来越多地发出API调用以改变现实系统,但许多现有架构直接将随机模型输出传递给执行层。我们主张这种耦合会带来安全风险,因为模型正确性、上下文意识和对齐性在执行时无法保证。我们引入主权代理循环(SAL),一种控制平面架构,其中模型发出带有理由的结构化意图,控制平面在执行前验证这些意图与真实系统状态和政策一致。SAL结合了限制模型访问身份敏感状态的混淆膜,以及用于可审计性和回放的加密链接证据链。我们正式化了SAL,并证明在所陈述的假设下,它提供了政策有界执行、身份隔离和确定性回放。在OpenKedge云基础设施原型中,SAL在政策层阻止了93%的危险意图,拒绝剩余7%通过一致性检查,防止了不安全的执行,并增加了12.4毫秒的中位延迟。

英文摘要:

Large language model (LLM) agents increasingly issue API calls that mutate real systems, yet many current architectures pass stochastic model outputs directly to execution layers. We argue that this coupling creates a safety risk because model correctness, context awareness, and alignment cannot be assumed at execution time. We introduce Sovereign Agentic Loops (SAL), a control-plane architecture in which models emit structured intents with justifications, and the control plane validates those intents against true system state and policy before execution. SAL combines an obfuscation membrane, which limits model access to identity-sensitive state, with a cryptographically linked Evidence Chain for auditability and replay. We formalize SAL and show that, under the stated assumptions, it provides policy-bounded execution, identity isolation, and deterministic replay. In an OpenKedge prototype for cloud infrastructure, SAL blocks 93% of unsafe intents at the policy layer, rejects the remaining 7% via consistency checks, prevents unsafe executions in our benchmark, and adds 12.4 ms median latency.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑