Agent systems increasingly execute rather than advise. When an AI agent queries regulated data, invokes effectful tools, and mutates persistent state, correctness is not captured by whether a terminal output looks plausible. The operative questions are whether each step was authorized under a contract, whether the recorded history is tamper-evident, and whether the trajectory can be reconstructed deterministically. We formalize this as runtime proof of execution. An execution is a triple $x = (C, T, R)$: a contract $C$, an Execution Causal Event Stream (ECES) $T$, and a replay context $R$. A well-formedness predicate and five validator-checkable invariants form the PoE validity predicate. Five semantic guarantees describe authorization, path compliance, null effect on deny, history integrity, and replayability. We prove soundness under explicit cryptographic and deployment assumptions: any PPT adversary that produces a PoE-valid execution violating a semantic guarantee yields a signature forgery, a hash collision, or a quantified deployment-failure event. The Prime Execution Model (PEM) separates planning, enforcement, effect, and recordkeeping into distinct authority planes; a lemma reduces trace completeness to Effector-exclusive credentialing. An Execution Attestation Certificate is issued only when PoE = 1. In a single-node TypeScript prototype, PoE adds approximately 2.7 ms on a minimal flow and 4.4% overhead on concurrent batch workloads; a standard eight-event trace compresses to approximately 1.1 KB; injected Gateway-bypass and trace-mutation attacks are rejected. PoE does not replace consensus, TEEs, or zkVMs; it binds authorization, effect, history, and replay into a single runtime-checkable object so that governed execution becomes attestable under contract.
Multi-Aspect Runtime Verification for Simulation-Based V&V of LLM-Enabled Autonomous Agents
Clone Systems(Clone Systems公司);International Hellenic University(国际希腊大学);University of Thessaly(色萨利大学);University of Peloponnese(伯罗奔尼撒大学);Aristotle University of Thessaloniki(塞萨洛尼基亚里士多德大学)
Formal Runtime Verification for Tool-Using LLM Agents: An Offline Same-Benchmark Study on AgentDojo and STAC
Clone Systems(Clone Systems公司);International Hellenic University(国际希腊大学);University of Thessaly(色萨利大学);Aristotle University of Thessaloniki(塞萨洛尼基亚里士多德大学)
Hybrid Hierarchical Runtime Verification for Edge-IoT Security: Combining MonPoly and RTLola
Clone Systems(Clone Systems公司);International Hellenic University(国际希腊大学);Aristotle University of Thessaloniki(塞萨洛尼基亚里士多德大学);University of Thessaly(色萨利大学)