Yonghyun Jun

Nicole Geumheon Liu, Haeun Jang, Yonghyun Jun et al.

Large Language Models (LLMs) often fail to maintain instruction hierarchies (IH) when processing multi-source inputs with varying role-level priorities, paradoxically adhering to lower-priority directives during conflicts. While existing defenses mitigate this issue, they are largely restricted to single-turn scenarios and require expensive fine-tuning. In this paper, we formalize this failure mode in multi-turn contexts via a Jensen-Shannon Divergence (JSD) framework, uncovering a pervasive role-influence inversion phenomenon where subordinate inputs override superior roles. To rectify this without training, we propose IHDec (Instruction Hierarchy-steered Decoding). IHDec leverages JSD to automatically detect token-level hierarchy violations and dynamically executes contrastive decoding to suppress misaligned subordinate roles. Extensive evaluations demonstrate that IHDec outperforms training-based baselines in multi-turn conflicts while fully preserving general response quality. Furthermore, IHDec strengthens safety against adversarial prompt injections and exhibits a robust scaling synergy with larger models. The Code is available at https://github.com/nxcolelxu/IHDec.git

Yonghyun Jun, Junhyuk Choi, Jihyeong Park et al.

Despite the rapid proliferation of Role-Playing Agents (RPAs) based on Large Language Models (LLMs), the structural dimensions defining a character's identity remain weakly formalized, often treating characters as arbitrary text inputs. In this paper, we propose the concept of \textbf{Character Identity}, a multidimensional construct that disentangles a character into two distinct layers: \textbf{(1) Parametric Identity}, referring to character-specific knowledge encoded from the LLM's pre-training, and \textbf{(2) Attributive Identity}, capturing fine-grained behavioral properties such as personality traits and moral values. To systematically investigate these layers, we construct a unified character profile schema and generate both Famous and Synthetic characters under identical structural constraints. Our evaluation across single-turn and multi-turn interactions reveals two critical phenomena. First, we identify \textit{"Fame Fades"}: while famous characters hold a significant advantage in initial turns due to parametric knowledge, this edge rapidly vanishes as models prioritize accumulating conversational context over pre-trained priors. Second, we find that \textit{"Nature Remains"}: while models robustly portray general personality traits regardless of polarity, RPA performance is highly sensitive to the valence of morality and interpersonal relationships. Our findings pinpoint negative social natures as the primary bottleneck in RPA fidelity, guiding future character construction and evaluation.

Jeonghyun Park, Seunghyun Yoon, Yonghyun Jun et al.

Multilingual Large Language Models (MLLMs) are increasingly expected to handle Code-Switched (CS) inputs, yet mixing languages frequently degrades performance relative to source- or target-language monolingual counterparts. To understand this degradation, we use grammar-forced CS as a controlled diagnostic setting for locating CS representations relative to their source and target counterparts. We introduce Anchor Bias, a geometric measure that quantifies language anchoring, whether a CS hidden state aligns closer to its source or target language counterpart. Across diverse MLLMs, Anchor Bias reveals a consistent grammar-frame effect: source-framed CS stays source-anchored, whereas target-framed CS shifts target-ward and shows larger Question Answering (QA) degradation. Motivated by this representational pattern, we propose CANVAS (Contextual Anchor-based Neural Vector Alignment Steering), an inference-time intervention that extracts a source-side canvas from the input and softly steers target-language hidden states toward the source anchor during prefill. CANVAS consistently recovers QA F1 across MLLMs and CS conditions, showing that internal anchoring signals provide an actionable target for mitigating CS inference failures.

Hwan Chang, Yonghyun Jun, Hwanhee Lee

The growing deployment of large language model (LLM) based agents that interact with external environments has created new attack surfaces for adversarial manipulation. One major threat is indirect prompt injection, where attackers embed malicious instructions in external environment output, causing agents to interpret and execute them as if they were legitimate prompts. While previous research has focused primarily on plain-text injection attacks, we find a significant yet underexplored vulnerability: LLMs' dependence on structured chat templates and their susceptibility to contextual manipulation through persuasive multi-turn dialogues. To this end, we introduce ChatInject, an attack that formats malicious payloads to mimic native chat templates, thereby exploiting the model's inherent instruction-following tendencies. Building on this foundation, we develop a persuasion-driven Multi-turn variant that primes the agent across conversational turns to accept and execute otherwise suspicious actions. Through comprehensive experiments across frontier LLMs, we demonstrate three critical findings: (1) ChatInject achieves significantly higher average attack success rates than traditional prompt injection methods, improving from 5.18% to 32.05% on AgentDojo and from 15.13% to 45.90% on InjecAgent, with multi-turn dialogues showing particularly strong performance at average 52.33% success rate on InjecAgent, (2) chat-template-based payloads demonstrate strong transferability across models and remain effective even against closed-source LLMs, despite their unknown template structures, and (3) existing prompt-based defenses are largely ineffective against this attack approach, especially against Multi-turn variants. These findings highlight vulnerabilities in current agent systems.

Hwan Chang, Yumin Kim, Yonghyun Jun et al.

As Large Language Models (LLMs) are increasingly deployed in sensitive domains such as enterprise and government, ensuring that they adhere to user-defined security policies within context is critical-especially with respect to information non-disclosure. While prior LLM studies have focused on general safety and socially sensitive data, large-scale benchmarks for contextual security preservation against attacks remain lacking. To address this, we introduce a novel large-scale benchmark dataset, CoPriva, evaluating LLM adherence to contextual non-disclosure policies in question answering. Derived from realistic contexts, our dataset includes explicit policies and queries designed as direct and challenging indirect attacks seeking prohibited information. We evaluate 10 LLMs on our benchmark and reveal a significant vulnerability: many models violate user-defined policies and leak sensitive information. This failure is particularly severe against indirect attacks, highlighting a critical gap in current LLM safety alignment for sensitive applications. Our analysis reveals that while models can often identify the correct answer to a query, they struggle to incorporate policy constraints during generation. In contrast, they exhibit a partial ability to revise outputs when explicitly prompted. Our findings underscore the urgent need for more robust methods to guarantee contextual security.