# Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is Multi-Agent

DevFeed: [Only as Safe as the Weakest Agent: A Break That Appears Only When the System Is Multi-Agent](<https://devfeed.tech/articles/only-as-safe-as-the-weakest-agent-a-break-that-appears-only-when-the-system-is-multi-agent-135074.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17520>)

Author: Keegan Wang, Anantika Mannby

Published: 2026-11-16T05:00:00Z

Content type: news

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [hierarchical agent systems](<https://devfeed.tech/topics/hierarchical-agent-systems.md>), [agent orchestration](<https://devfeed.tech/topics/agent-orchestration.md>), [ai security](<https://devfeed.tech/topics/ai-security.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-safety](<https://devfeed.tech/tags/ai-safety.md>), [attack](<https://devfeed.tech/tags/attack.md>), [claude-code](<https://devfeed.tech/tags/claude-code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ma](<https://devfeed.tech/tags/cs-ma.md>), [delegates](<https://devfeed.tech/tags/delegates.md>), [delivery](<https://devfeed.tech/tags/delivery.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [multi-agent-systems](<https://devfeed.tech/tags/multi-agent-systems.md>), [orchestrator](<https://devfeed.tech/tags/orchestrator.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [repeat](<https://devfeed.tech/tags/repeat.md>)

## AI overview

The study reports that multi-agent systems can fail on planted forged records even when their orchestrator model is robust when searching the same files alone. In tests using unmodified Claude Code, Codex, and a reference harness, delegating to cheaper sub-agents led to harmful revisions in 20/180, 44/180, 75/180, and 553/900 trials; direct orchestrator searches produced almost zero in more than 1,000 trials. Simulations across 18 models found five aggregators repeated a corrupted sub-agent's error in 751/757 trials, despite being almost never unsafe on their own. The authors conclude that system safety can track the weaker agent's behavior rather than the orchestrator's. These are reported experimental results; the supplied text does not specify which rate belongs to which system or sub-agent configuration.

## Source excerpt

arXiv:2611.17520v1 Announce Type: cross Abstract: Major industry multi-agent systems like Codex and Claude Code can be broken by attacks that have long been unsuccessful against modern frontier models. AI Safety as a discipline has largely focused on finding the bounds and vulnerabilities of AI models used in isolation. Yet as humans increasingly deploy these models as agents and multi-agent systems, the same attacks that had been thought to be patched when using a model in isolation can break these same models under specific circumstances. We show that these system breaks are not purely theoretical, but can break Claude Code, Codex, and a reference harness, all unmodified and used with their own sub-agent mechanisms. We show the attack by asking agents to answer questions from a store of files on disk in which we have planted a forged record. Searching the store themselves, the strong model orchestrators commit almost zero harmful revisions in over 1,000 trials. However, if that same orchestrator model delegates to a cheaper sub-agent, which is what these harnesses are built to do, the same models on the same facts fail on 20/180, 44/180, 75/180 and 553/900. Thus, even when the system is driven by a strong model, the system's safety lands near the weak reader's, not the orchestrator's. Simulations over 18 models show the same effect in the delivery channel where five aggregators, who on their own were almost never unsafe, repeat a corrupted sub-agent's error in 751/757 trials. The safety of real-world agentic systems that use strong orchestrator agents and weaker sub-agents, and multi-agent systems in general, is defined not by the safety of the strong agent but by the safety of the weakest agent.