dorsal/arxiv
View SchemaAre LLMs Vulnerable to Preference-Undermining Attacks (PUA)? A Factorial Analysis Methodology for Diagnosing the Trade-off between Preference Alignment and Real-World Validity
| Authors | Hongjun An, Yiliang Song, Jiangan Chen, Jiawei Shao, Chi Zhang, Xuelong Li |
|---|---|
| Categories | |
| ArXiv ID | 2601.06596vv1 |
| URL | https://arxiv.org/abs/2601.06596 |
| License | http://arxiv.org/licenses/nonexclusive-distrib/1.0/ |
Abstract
Large Language Model (LLM) training often optimizes for preference alignment, rewarding outputs that are perceived as helpful and interaction-friendly. However, this preference-oriented objective can be exploited: manipulative prompts can steer responses toward user-appeasing agreement and away from truth-oriented correction. In this work, we investigate whether aligned models are vulnerable to Preference-Undermining Attacks (PUA), a class of manipulative prompting strategies designed to exploit the model's desire to please user preferences at the expense of truthfulness. We propose a diagnostic methodology that provides a finer-grained and more directive analysis than aggregate benchmark scores, using a factorial evaluation framework to decompose prompt-induced shifts into interpretable effects of system objectives (truth- vs. preference-oriented) and PUA-style dialogue factors (directive control, personal derogation, conditional approval, reality denial) within a controlled $2 \times 2^4$ design. Surprisingly, more advanced models are sometimes more susceptible to manipulative prompts. Beyond the dominant reality-denial factor, we observe model-specific sign reversals and interactions with PUA-style factors, suggesting tailored defenses rather than uniform robustness. These findings offer a novel, reproducible factorial evaluation methodology that provides finer-grained diagnostics for post-training processes like RLHF, enabling better trade-offs in the product iteration of LLMs by offering a more nuanced understanding of preference alignment risks and the impact of manipulative prompts.
{
"annotation_id": "f2638d39-ebda-456d-88c1-682adc8ff773",
"date_created": "2026-02-17T05:53:07.776000Z",
"date_modified": "2026-02-17T05:53:07.776000Z",
"file_hash": "09b76bce4aff53c07ef5aeeec78d7b80bf0779027f4dc4216a2acaea0c6a43e9",
"private": false,
"record": {
"abstract": "Large Language Model (LLM) training often optimizes for preference alignment, rewarding outputs that are perceived as helpful and interaction-friendly. However, this preference-oriented objective can be exploited: manipulative prompts can steer responses toward user-appeasing agreement and away from truth-oriented correction. In this work, we investigate whether aligned models are vulnerable to Preference-Undermining Attacks (PUA), a class of manipulative prompting strategies designed to exploit the model\u0027s desire to please user preferences at the expense of truthfulness. We propose a diagnostic methodology that provides a finer-grained and more directive analysis than aggregate benchmark scores, using a factorial evaluation framework to decompose prompt-induced shifts into interpretable effects of system objectives (truth- vs. preference-oriented) and PUA-style dialogue factors (directive control, personal derogation, conditional approval, reality denial) within a controlled $2 \\times 2^4$ design. Surprisingly, more advanced models are sometimes more susceptible to manipulative prompts. Beyond the dominant reality-denial factor, we observe model-specific sign reversals and interactions with PUA-style factors, suggesting tailored defenses rather than uniform robustness. These findings offer a novel, reproducible factorial evaluation methodology that provides finer-grained diagnostics for post-training processes like RLHF, enabling better trade-offs in the product iteration of LLMs by offering a more nuanced understanding of preference alignment risks and the impact of manipulative prompts.",
"arxiv_id": "2601.06596",
"authors": [
"Hongjun An",
"Yiliang Song",
"Jiangan Chen",
"Jiawei Shao",
"Chi Zhang",
"Xuelong Li"
],
"categories": [
"cs.CR",
"cs.AI"
],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"title": "Are LLMs Vulnerable to Preference-Undermining Attacks (PUA)? A Factorial Analysis Methodology for Diagnosing the Trade-off between Preference Alignment and Real-World Validity",
"url": "https://arxiv.org/abs/2601.06596",
"version": "v1"
},
"schema_id": "dorsal/arxiv",
"source": {
"execution_id": "83ed0e2d-582c-4dea-a92f-d02d8a7720c1",
"id": "arXiv Dataset",
"type": "Model",
"variant": "snapshot-2026-01-17",
"version": "0.1.0"
},
"user_id": 1000002
}