Research
Scalable Hierarchical Attention Transformers for Multi-Turn Jailbreak Detection in Long Conversations
Publication date not yet evaluated · Added arxiv.org
Page published
Coverage timeline
Single-source research — one report is available.
Why it matters
Multi-turn jailbreaks evade turn-level moderation, so a scalable conversation-level detector gives defenders a practical way to catch attacks that distribute malicious intent across a dialogue.
Researchers from Zscaler present a hierarchical attention transformer for detecting multi-turn jailbreaks that spread unsafe intent across a conversation via gradual escalation, reframing, and role manipulation. The two-level architecture encodes individual turns with a multilingual encoder then reasons over turn summaries with a lightweight conversation module, reporting an F1 of 0.9394 on a 14,038-conversation benchmark and outperforming a strong LLM baseline while halving false positives.