AgentRepEngine: Runtime Behavioral Enforcement for AI Agents — Production Implementation of the Agentic Trust Protocol (ATP) and Agentic Trust Graph (ATG)

Read full text Summarise with VIRAT
Authors: Rehan Masood
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

AgentRepEngine is the first production implementation of the Agentic Trust Protocol (ATP) and Agentic Trust Graph (ATG), providing runtime behavioral scoring and gateway-level enforcement for autonomous AI agents in regulated enterprise environments. This record establishes the implementation-layer citation anchor linking ATP/ATG foundational research to deployed enforcement infrastructure. Core capabilities:- Behavioral scoring at the API gateway (Kong plugin, RS256 JWT identity)- Velocity + z-score anomaly detection per agent session- Tamper-evident enforcement audit trail with hash chain verification- Structured reason objects on every enforcement decision (G-EXPLAIN)- Python SDK for LangChain and agent framework integration- 0.00% false positive rate on 100-scenario internal corpus- 100% slow-walk attack detection via HIGH_RISK VERIFY layer- Sub-millisecond call-level overhead (0.25ns on Linux) AgentRepEngine directly implements:- ATP identity semantics: signed JWT + RS256 + org JWKS (ATG identity nodes)- ATP enforcement semantics: TRUSTED / MONITORED / RESTRICTED / BLOCKED bands- ATG lineage tracking: lineage_hash per agent, sub-agent spawn depth- ATG conflict preservation: tamper-evident hash chain on enforcement_decisions GitHub: https://github.com/Rehanrana11/AgentRepEngine

Fields of Study & Subjects

Access Control and Trust Adversarial Robustness in Machine Learning Security and Verification in Computing

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
17 March 2026
Language
EN