InfoResearchPreprintLLM-specific
Jailbreak Detection on Large Language Models via Contrastive Prototype Structures
- Published
- Record updated
Summary
Researchers propose JDPI, a jailbreak detection method that builds harmful and benign prototypes from binary safety labels and classifies inputs by relative prototype distance, avoiding a separate downstream classifier. Across four benchmarks, contrastive fine-tuning raises detection accuracy by 9.1-18.0 percentage points, and JDPI exceeds the released Llama Guard 3-8B baseline by 0.173-0.310 in ROC-AUC with dataset-specific adaptation. In multi-turn replay, using accumulated user history yields higher conversation-level accuracy than using the current turn alone.
Topics
Related items
- InfoMeta’s Muse Charm might cost $300Same vendor · The Verge (AI)
- InfoJEV vs. LLMs: A Comparative Analysis for Phishing Email DetectionSame vendor · Arxiv (cs.CR + cs.CL + cs.LG)
- InfoNot even Sam Altman is safe from AI deepfakes on InstagramSame vendor · The Verge (AI)
- InfoProton’s Lumo, a privacy-focused alternative to ChatGPT and Claude, arrives on desktopSame vendor · TechCrunch (Security)
- InfoClaude’s Duet agent is a little neurotic and surprisingly helpfulSame vendor · The Verge (AI)