# SwarmSafe-Delta: Isolating Message Effects on LLM Agent Safety

DevFeed: [SwarmSafe-Delta: Isolating Message Effects on LLM Agent Safety](<https://devfeed.tech/articles/swarmsafe-delta-isolating-message-effects-on-llm-agent-safety-132199.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.16123>)

Author: Junru Zhu

Published: 2026-11-13T05:00:00Z

Content type: article

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>), [agent to agent communication](<https://devfeed.tech/topics/agent-to-agent-communication.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agent-communication](<https://devfeed.tech/tags/agent-communication.md>), [agent-safety](<https://devfeed.tech/tags/agent-safety.md>), [aggregate](<https://devfeed.tech/tags/aggregate.md>), [also](<https://devfeed.tech/tags/also.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [communication](<https://devfeed.tech/tags/communication.md>), [completion](<https://devfeed.tech/tags/completion.md>), [contribution](<https://devfeed.tech/tags/contribution.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ma](<https://devfeed.tech/tags/cs-ma.md>), [delivery](<https://devfeed.tech/tags/delivery.md>), [delta](<https://devfeed.tech/tags/delta.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [effect](<https://devfeed.tech/tags/effect.md>), [effects](<https://devfeed.tech/tags/effects.md>), [execution](<https://devfeed.tech/tags/execution.md>), [executor](<https://devfeed.tech/tags/executor.md>), [llm](<https://devfeed.tech/tags/llm.md>), [message](<https://devfeed.tech/tags/message.md>), [safety](<https://devfeed.tech/tags/safety.md>)

## AI overview

SwarmSafe-Delta measures how message delivery changes a tool-using agent's committed action by freezing upstream computation and comparing real, hidden, and format-matched placebo delivery across deterministic benign and risky policy states. In four models and constructed optional tasks, real delivery at N=4 yielded 21.4% risky-world violations, compared with 5.0% for placebo and 1.0% for hidden delivery. It also improved benign task success; in a separate regime requiring an upstream lookup, it raised safe completion by 74.2 percentage points over hidden delivery. Exploratory analyses found effects varied with whether messages advised a prerequisite check or direct execution. These are controlled contrasts under specified information dependencies, not estimates of naturally occurring failure rates or evidence that communication is generally harmful or protective.

## Source excerpt

arXiv:2611.16123v1 Announce Type: cross Abstract: When a tool-using LLM agent team violates policy, aggregate outcomes do not reveal whether an upstream message changed the executor's action. We introduce SwarmSafe-Delta, a paired executable protocol that freezes upstream computation, varies real, hidden, and format-matched placebo delivery, and scores a single committed action against deterministic benign and risky policy states under preassigned budgets. Across four models and constructed optional tasks, real delivery at $N=4$ produces 21.4% risky-world violations, versus 5.0% with placebo and 1.0% with hidden delivery (paired differences +16.4 and +20.3 percentage points). Real delivery also improves benign task success, illustrating that suppressing communication sacrifices utility. In a second regime constructed to require an upstream lookup route, real delivery improves safe completion by 74.2 points relative to hidden delivery. These effects quantify message-delivery contrasts under specified information dependencies, not naturally occurring failure rates. Exploratory action-level analyses show that the effect depends strongly on whether the upstream message recommends a prerequisite check or direct execution; model differences also largely track these message compositions. The contribution is a controlled, execution-grounded measurement protocol rather than a general claim that agent communication is harmful or protective.