In short: AI-Infra-Guard addresses the fragmented attack surface of AI agents through layer-specific security paradigms: rule-matching for infrastructure, LLM audits for protocols, and behavioral testing for agent conduct.
Researchers present AI-Infra-Guard, an open-source framework for securing AI agents across multiple architectural layers. The tool combines rule-based vulnerability detection, LLM-powered protocol auditing, and black-box red-teaming, thereby targeting a security gap in the rapidly growing AI infrastructure landscape.
The attack surface of modern AI agents spans four layers: infrastructure (model serving, agent platforms), protocols and tools (Model Context Protocol, MCP), agent behavior, and the underlying language models themselves. Previous security tools address these layers in isolation, resulting in gaps in practical protection coverage.
AI-Infra-Guard organizes red-teaming activities around the central insight that each layer requires a different detection paradigm. The framework integrates deterministic rule-matching across 75+ AI components and over 1,400 vulnerability rules, LLM-driven agent-based auditing of MCP servers and agent skill packages, as well as multi-turn black-box red-teaming against agent behavior itself. Additionally, it includes a jailbreak testing harness with over 26 attack operators across sixteen datasets.
A distinctive aspect of the approach is supply-chain verification of agent skills — extensions that increasingly expand the functionality of AI agents and thus create new attack vectors. According to the authors, this makes the framework the only open-source solution that comprehensively covers all layers. The system is available to the community as open source to establish a common foundation for agent security.
Source: arxiv.org · Published 29 June 2026
Lumi AI News — AI-assisted curation in accordance with Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.