LGCLCRMay 30

Same Payload, Different Channel: Measuring Trust Asymmetry in Tool-Using Language Models

arXiv:2606.0056647.1h-index: 10
AI Analysis

For developers of tool-using LLMs, this reveals a systematic, channel-dependent blind spot in safety mechanisms that current models fail to address.

The paper introduces the Safety Asymmetry Score (SAS) to measure how LLMs' susceptibility to adversarial content varies by delivery channel (user message, tool metadata, tool output). Across 6 models, agent-native models are more vulnerable via tool descriptions, while general-purpose models show the opposite; the asymmetry inverts for tool outputs.

As language models take on agentic roles that span calling external APIs, reading tool outputs, and acting on instructions embedded in third-party content, their attack surface expands well beyond what users type. Whether a model treats a malicious instruction the same way regardless of where it arrives has not been systematically studied. We introduce the Safety Asymmetry Score (SAS), which measures how much a model's susceptibility to adversarial content shifts depending on whether that content arrives in the user message, tool metadata, or tool output, using matched payload pairs that keep the malicious text identical and vary only the context of delivery. Evaluated across 6 production LLMs and three attack families, we find a consistent and informative asymmetry: agent-native models are substantially more vulnerable when adversarial content arrives via tool descriptions than via user messages, while general-purpose models show the reverse. This asymmetry further inverts when the same content is delivered through tool outputs rather than descriptions, suggesting models implicitly treat tool metadata as trusted instructions and tool results as ordinary data. A mechanistic study on Llama 3.3 70B reveals that the safety-relevant representation is causally present at mid-to-late network depths but non-linearly encoded, explaining why linear probes fail to detect it. These findings expose a systematic, channel-dependent blind spot in how current tool-using models handle adversarial content.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes