Ellipsoid Control: A White-list Jailbreak Defense via Benign Latent Modeling
- 1 January 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Information Forensics and Security
- Vol. 21, 1
Abstract
No abstract availableAll Related Versions
Funding Information
- Australian Research Council (ARC), Australia, (LP220100453)
This publication has 8 references indexed in Scilit:
- HSF: Defending against Jailbreak Attacks with Hidden State Filtering2025
- Refusal in Language Models Is Mediated by a Single Direction2024
- Gradient Cuff: Detecting Jailbreak Attacks on Large Language Models by Exploring Refusal Loss Landscapes2024
- GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysis2024
- Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena2023
- Enhancing Chat Language Models by Scaling High-quality Instructional Conversations2023
- Iron: Private Inference on Transformers2022
- Efficient and Privacy-Enhanced Federated Learning for Industrial Artificial IntelligenceIEEE Transactions on Industrial Informatics, 2019