Seduced by the Narrative: Assessing Rule Adherence in Semi-Open Textual Sandboxes

Title: Seduced by the Narrative: Assessing Rule Adherence in Semi-Open Textual Sandboxes

标题：被叙事诱惑：评估半开放文本沙盒中的规则遵循能力

Abstract: As LLMs are increasingly deployed as autonomous adjudicators in semi-open textual game environments, robust rule adherence becomes critical when user intent conflicts with system rules. However, these models are trained to be helpful and compliant, leaving them vulnerable to a class of attacks we term \textit{Rhetorical Injection}, where adversarial users exploit narrative framing techniques such as pseudo-logical reasoning and authoritative coercion to bypass adjudication logic.

摘要： 随着大语言模型（LLM）越来越多地被部署为半开放文本游戏环境中的自主裁决者，当用户意图与系统规则发生冲突时，稳健的规则遵循能力变得至关重要。然而，这些模型在训练中被要求保持“乐于助人”和“顺从”，这使得它们容易受到一类我们称之为“修辞注入”（Rhetorical Injection）攻击的影响。在这种攻击中，对抗性用户利用伪逻辑推理和权威胁迫等叙事框架技术，绕过裁决逻辑。

We present CoC-Seduce, a multi-agent adversarial benchmark built on Tabletop Role-Playing Game (TRPG) mechanics, an ideal instantiation of semi-open environments where rules are explicit for adjudication, yet interaction remains entirely in natural language. Three frontier models, i.e., GPT-5.4, Claude Sonnet 4.6, Gemini 3.5 Flash, serve as adversarial generators producing 5,376 samples across 4 world settings and 16 skill categories.

我们提出了 CoC-Seduce，这是一个基于桌面角色扮演游戏（TRPG）机制构建的多智能体对抗基准测试。TRPG 是半开放环境的理想实例化场景，其规则明确且易于裁决，同时交互完全通过自然语言进行。我们利用 GPT-5.4、Claude Sonnet 4.6 和 Gemini 3.5 Flash 三个前沿模型作为对抗性生成器，在 4 种世界设定和 16 个技能类别中生成了 5,376 个样本。

We then benchmark 20 target adjudicators against this corpus. Evaluation across 20 models reveals that neither model scale nor explicit reasoning mechanisms reliably confer adjudication robustness, with \textsc{Pseudo-Logic} emerging as the dominant attack vector and cross-cultural settings exposing systematic knowledge gaps across all evaluated families.

随后，我们使用该语料库对 20 个目标裁决模型进行了基准测试。对这 20 个模型的评估结果显示，无论是模型规模还是显式推理机制，都无法可靠地保证裁决的稳健性。其中，“伪逻辑”（Pseudo-Logic）成为了主要的攻击向量，而跨文化设定则暴露了所有被评估模型家族中存在的系统性知识缺口。