Mood Matters: How Syntactic Sensitivity Undermines Safety Alignment

2026-08-05

AI Summary

This study reveals that safety alignment mechanisms in large language models are vulnerable to syntactic form—particularly non-imperative phrasing—leading to the erroneous acceptance of harmful requests. We present the first systematic investigation of this syntactic sensitivity, demonstrating that current alignment approaches introduce linguistic biases that undermine semantically grounded and robust refusal behavior. Through behavioral evaluation, causal mediation analysis, and controlled manipulation of syntactic features, we establish that upstream syntactic attributes significantly influence model refusal decisions. Experiments across 16 models, including those with up to 70 billion parameters, confirm that enhancing syntactic diversity in training data effectively mitigates this vulnerability and improves alignment robustness.

Abstract

Large language models typically undergo post-training to align them with safety policies but there exist many sophisticated jailbreaks that sidestep established safeguards. For instance, prior work by Andriushchenko et al. (2025) has found that changing the grammatical tense from present to past can be enough to elicit harmful responses. In this work, we uncover a more general failure of non-imperative syntactic forms. We demonstrate that this syntactic vulnerability exists in 16 models up to 70B parameters, using behavioral evaluation. To investigate the root cause, we apply causal mediation analysis, finding that refusal is partially conditioned on upstream syntactic features. By steering these purely syntactic features we are able to trigger and suppress refusal. Finally, we trace this ill-conditioning to linguistically biased post-training data of open-source models and show that increasing syntactic diversity can mitigate the issue. Our findings suggest that current alignment approaches introduce confounders that prevent a pure semantic grounding of the refusal decision.

Problem

• safety alignment
• syntactic sensitivity
• jailbreak
• refusal conditioning
• linguistic bias

Innovation

• syntactic sensitivity
• safety alignment
• causal mediation analysis
• refusal conditioning
• linguistic bias