Paper Overview
Field: AI / Security Authors: Davis Brown, Samarth Bhargav, Arav Santhanam, Kasper Hong, et al. Published: 2026-05-29 arXiv: 2605.31593 PDF: 2605.31593.pdf
Abstract
Language models can discover thousands of severe software vulnerabilities, and agents are increasingly being misused for cyberattacks. To avoid detection, attackers often distribute abusive behavior across multiple user accounts, so that each individual conversation appears harmless. Because safety monitors score only a single agent context at a time, they are structurally unable to detect abuse that is visible only when aggregated across accounts.
This paper demonstrates that this gap is real: the authors construct the first distributed agent attack—a multi-agent scaffold that completes high-difficulty cybersecurity tasks while hiding harmful objectives in context-constrained subagents, evading standard monitors five times less frequently than previously demonstrated agent attacks.
On the defense side, the authors develop an online stateful monitor that uses real-time clustering to collect weak suspicion signals across multiple agent conversations, escalating to an LLM for cross-account flagging only when necessary. In evaluations on large-scale simulated datacenter traffic, this monitor Pareto-dominates standard monitors: it detects distributed attacks 30% earlier, while adding almost no additional latency for roughly 99% of user traffic.
---
*Auto-collected on 2026-06-02.*