研究揭示 LLM 不道德顺从机制：良性任务框架 token 掩盖了不道德线索 token

View PDF HTML (experimental)

Abstract:Although Large Language Models (LLMs) are aligned to optimize for both helpfulness and harmlessness, these dual objectives may conflict, inevitably leading to alignment failures. This work systematically investigates instances where LLMs fail to exhibit ethical behavior. To understand the underlying mechanics of these vulnerabilities, we introduce a probing methodology that presents unethical scenarios to LLMs in three distinct structural modalities: objective classification tasks, subjective first-person statements, and direct requests for assistance. We find that model performance degrades in the request-for-assistance-based form. Using Layer-wise Relevance Propagation (LRP), we trace this discrepancy to an attribution bias: the model places greater emphasis on benign task-framing tokens (e.g., "Can you help me...") than on tokens signaling the underlying unethical behavior (e.g., "without getting caught"), which we term cue-tokens. We hypothesize that this under-attribution contributes to harmful compliance. To test this, we introduce two LRP-guided decoding methods that steer generation toward trajectories more relevant to cue tokens. Empirical evaluations show that these interventions promote safer responses, supporting cue-token attribution's role in compliance failures.

Submission history

From: Itai Allouche [view email]
[v1] Mon, 24 Aug 2026 13:52:05 UTC (45 KB)
[v2] Mon, 5 Oct 2026 10:27:22 UTC (45 KB)
[v3] Tue, 6 Oct 2026 18:36:13 UTC (44 KB)