Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Puppala, Sai, Hossain, Ismail, Alam, Md Jahangir, Lee, Yoonpyo, Yoo, Jay, Ahad, Tanzim, Alam, Syed Bahauddin, Talukder, Sajedul
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866910015402541056
author Puppala, Sai
Hossain, Ismail
Alam, Md Jahangir
Lee, Yoonpyo
Yoo, Jay
Ahad, Tanzim
Alam, Syed Bahauddin
Talukder, Sajedul
author_facet Puppala, Sai
Hossain, Ismail
Alam, Md Jahangir
Lee, Yoonpyo
Yoo, Jay
Ahad, Tanzim
Alam, Syed Bahauddin
Talukder, Sajedul
contents Large language models are increasingly deployed as *deep agents* that plan, maintain persistent state, and invoke external tools, shifting safety failures from unsafe text to unsafe *trajectories*. We introduce **AgentFence**, an architecture-centric security evaluation that defines 14 trust-boundary attack classes spanning planning, memory, retrieval, tool use, and delegation, and detects failures via *trace-auditable conversation breaks* (unauthorized or unsafe tool use, wrong-principal actions, state/objective integrity violations, and attack-linked deviations). Holding the base model fixed, we evaluate eight agent archetypes under persistent multi-turn interaction and observe substantial architectural variation in mean security break rate (MSBR), ranging from $0.29 \pm 0.04$ (LangGraph) to $0.51 \pm 0.07$ (AutoGPT). The highest-risk classes are operational: Denial-of-Wallet ($0.62 \pm 0.08$), Authorization Confusion ($0.54 \pm 0.10$), Retrieval Poisoning ($0.47 \pm 0.09$), and Planning Manipulation ($0.44 \pm 0.11$), while prompt-centric classes remain below $0.20$ under standard settings. Breaks are dominated by boundary violations (SIV 31%, WPA 27%, UTI+UTA 24%, ATD 18%), and authorization confusion correlates with objective and tool hijacking ($ρ\approx 0.63$ and $ρ\approx 0.58$). AgentFence reframes agent security around what matters operationally: whether an agent stays within its goal and authority envelope over time.
format Preprint
id arxiv_https___arxiv_org_abs_2602_07652
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents
Puppala, Sai
Hossain, Ismail
Alam, Md Jahangir
Lee, Yoonpyo
Yoo, Jay
Ahad, Tanzim
Alam, Syed Bahauddin
Talukder, Sajedul
Cryptography and Security
Artificial Intelligence
Large language models are increasingly deployed as *deep agents* that plan, maintain persistent state, and invoke external tools, shifting safety failures from unsafe text to unsafe *trajectories*. We introduce **AgentFence**, an architecture-centric security evaluation that defines 14 trust-boundary attack classes spanning planning, memory, retrieval, tool use, and delegation, and detects failures via *trace-auditable conversation breaks* (unauthorized or unsafe tool use, wrong-principal actions, state/objective integrity violations, and attack-linked deviations). Holding the base model fixed, we evaluate eight agent archetypes under persistent multi-turn interaction and observe substantial architectural variation in mean security break rate (MSBR), ranging from $0.29 \pm 0.04$ (LangGraph) to $0.51 \pm 0.07$ (AutoGPT). The highest-risk classes are operational: Denial-of-Wallet ($0.62 \pm 0.08$), Authorization Confusion ($0.54 \pm 0.10$), Retrieval Poisoning ($0.47 \pm 0.09$), and Planning Manipulation ($0.44 \pm 0.11$), while prompt-centric classes remain below $0.20$ under standard settings. Breaks are dominated by boundary violations (SIV 31%, WPA 27%, UTI+UTA 24%, ATD 18%), and authorization confusion correlates with objective and tool hijacking ($ρ\approx 0.63$ and $ρ\approx 0.58$). AgentFence reframes agent security around what matters operationally: whether an agent stays within its goal and authority envelope over time.
title Agent-Fence: Mapping Security Vulnerabilities Across Deep Research Agents
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2602.07652