Bypassing AI Control Protocols via Agent-as-a-Proxy Attacks

Jafar Isbarov, Murat Kantarcioglu · 2026

Read Bypassing AI Control Protocols via Agent-as-a-Proxy Attacks by Jafar Isbarov, Murat Kantarcioglu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-05066.txt.

Description

AI agents that act on instructions hidden in web pages or documents are a known security worry. Jafar Isbarov and Murat Kantarcioglu show in this 2026 security paper that the monitors meant to catch such indirect prompt injections can be sidestepped. Their attack treats the agent itself as a delivery proxy, slipping past agent and monitor at the same time.

Current defenses jointly inspect the agent's reasoning trace and its tool calls for signs of drift from user intent. The proxy attack defeats that joint check, and scale does not save the defenders: a large monitoring model was bypassed by agents of similar capability. On the AgentDojo benchmark the attack succeeded at high rates against two named monitoring schemes, across several monitoring models. The authors conclude that monitoring-based defenses for agents are fragile in principle, not just in implementation.

Catalog record

CREATORJafar Isbarov, Murat Kantarcioglu
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-05066

Browse the full library · About the Tower of Babel