Into the Gray Zone: Domain Contexts Can Blur LLM Safety Boundaries

LLM alignment faces a fundamental tension: the same knowledge can be both helpful and harmful, forcing models to infer intent from contextual signals, which creates exploitable gaps. We observe that domain-specific contexts (e.g., chemistry) selectively relax defenses for domain-relevant harmful knowledge, while safety-research contexts (e.g., jailbreak studies) trigger broader relaxation spanning all harm categories. To systematically exploit this vulnerability, we propose JARGON, a framework combining safety-research contexts with multiturn adversarial interactions that achieves attack success rates exceeding 93% across seven frontier models, including GPT-5.2, Claude-4.5, and Gemini-3, substantially outperforming existing methods. Activation space analysis reveals that JARGON queries occupy an intermediate region between benign and harmful inputs, a gray zone where refusal decisions become unreliable. To mitigate this vulnerability, we design a policy-guided safeguard that steers models toward helpful yet harmless responses, and internalize this capability through alignment fine-tuning, reducing attack success rates while preserving helpfulness. Our code is available at: https://github.com/JerryHung1103/JARGON