Jingyu Hua

ECHO: Replacing the Compliance Proxy with Response Harmfulness in Optimization-Based Jailbreak Attack

This work proposes ECHO (Evaluation-Conditioned Harmfulness Optimization), which replaces the compliance proxy with response-level, continuous harmfulness scores from HarmSig, a 1.7B regressor grounded in actual harmful capability.

Adaptive Location Privacy in Location-Based Services: Service-Specific Perturbation for Balancing Security and Utility

Secret Stealing Attacks on Local LLM Fine-Tuning through Supply-Chain Model Code Backdoors

This work introduces a deterministic full-chain memorization mechanism that locks onto token-level secrets in dynamic computation flows via online tensor-rule matching, and leverages value-gradient decoupling to stealthily inject attack gradients, overcoming gradient drowning to force model memorization.

Mitigating Database Leakage in RAG Systems with Keyword-Grounded Fact Substitution

KFS-RAG is proposed, a defense that mitigates information leakage by reformulating the retrieved context by identifying a small set of influential keywords from the retrieved context via an attention rollout plus a causal perturbation mechanism.

PURPOSE: Poisoning Conflict Resolution in RAG via Proxy-Fact-Grounded Updates

PURPOSE is proposed, a strict black-box poisoning attack that reframes the injection as an update that minimizes conflict, rather than as a counter-claim, and identifies non-contradicting injection as a practical mode to enhance poisoning attack.