This analysis brings to light two novel categories of attacks in the context assembly of real-world harnesses: MessageRole Context Privilege Escalation (M-CPE), which occurs when attacker-controlled content originating from a low-privileged context is incorporated into a higher-privileged message role.
Abstract
Real-world, high-profile AI agent harnesses often rely on vendor-proprietary or opaque designs for context assembly, leaving the sources and underlying logic of assembled context poorly understood and the resulting security risks largely unexplored. In this paper, we present the first systematic analysis of context assembly designs in real-world AI agent harnesses. We study and uncover how an agent harness is designed to collect and assemble context from diverse sources, and identify a set of practical attack vectors arising from these designs. Our analysis brings to light two novel categories of attacks in the context assembly of real-world harnesses: (1) MessageRole Context Privilege Escalation (M-CPE), which occurs when attacker-controlled content originating from a low-privileged context is incorporated into a higher-privileged message role. (2) Cross-Scope Context Privilege Escalation (X-CPE), which occurs when attacker-controlled content persists beyond the context in which it was introduced. We performed a systemic security analysis of the CPE attacks against 12 real-world agent harnesses, including Claude Code and Codex. The resulting consequences include full agent compromise, remote code execution, denial of service, and manipulated tool or skill invocations, etc.
This work constructs a concrete prompt-injection attack that succeeds despite all constituent agents individually operating CaMeL, and introduces multi-CaMeL, an agent-to-agent communication protocol that preserves provenance across agent boundaries by separating trusted natural-language instructions from untrusted dat...
James Peters-Gill, Avi Semler, Henning Bartsch et al.· 3 citations
ProtoArena is introduced, a black-box, multi-component, dynamic benchmark with 80 cross-component vulnerability-exploitation environments spanning 11 protocol and application domains and 7 attack categories and success with limited black-box information remains low and varies across domains and attack categories.
Fa-Sheng Miao, Zhao-Yu Hu, Xiang Li et al.· 0 citations
Modern Computer Use Agents (CUAs) directly interact with graphical user interfaces and execute third-party web tools, exposing them to indirect prompt injection across every rendered page and tool response. While the Dual-LLM pattern is the primary system-level architecture offering formal security guarantees - using a...
Giulio Zingrillo, Hanna Foerster, Ilia Shumailov et al.· 0 citations
Enterprise AI agents now read tickets, documents, and web pages and then act on what they find through tool protocols such as the Model Context Protocol (MCP). Because an agent combines untrusted input with delegated privileges, it creates a path from attacker-influenced content to privileged action that earlier Zero T...
A. Solanke· World Journal of Advanced En...· 0 citations
Multi-agent LLM systems are increasingly deployed for complex, long-horizon tasks or emerge as a natural consequence of agents interacting in the wild. Yet they give rise to significant safety and security risks: the flexible protocols that enable task generalization also expose novel threats, from cascading prompt inj...
Ben Hagag, William L. Anderson, S. Chakraborty et al.· arXiv.org· 1 citation
Evidence concentrates on prompt/reasoning, memory, and tool-mediated attacks, predominantly in single-agent settings, within the selected studies, to guide more systematic, architecture-aware, and reproducible security evaluation of agentic AI.
Heewon Baek, Alsharif Abuadbba, Kristen Moore et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.