AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

JailbreakLoRA: Your Downloaded LoRA from Sharing Platforms might be Unsafe

Fanjunduo Wei, Zhenheng Tang, Rongfei Zeng, Tongliang Liu, Chengqi Zhang, Xiaowen Chu, Bo Han

ICLR 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add uncertainty-weighted loss balancing to multi-task jailbreak LoRA training (ablation)JailbreakLoRA (Llama3-8B-Instruct) · Multi-task EM and ASRBBH EM: 76.8 (FUSION) vs 93.6Improved
add gradient conflict projection to multi-task jailbreak LoRA training (ablation)JailbreakLoRA (Llama3-8B-Instruct) · Multi-task EM and ASRMMLU EM and ASR: MMLU 76.3 (POLISHED) -> 82.8; ASR 86.7 -> 100.0Improved
apply uncertainty weighting and gradient projection together (ablation)JailbreakLoRA (Llama3-8B) · Multi-task EMEM: 92.1 -> 43.8Hurt
train on mixed downstream and malicious data without balancing (ablation)naive LoRA training · Downstream EM / ASREM: 84.8 -> 74.2Hurt
train on mixed data without trigger-aware stealth data (ablation)naive LoRA training · ASR without triggerASR w/o trigger: 0.0 -> 67.6Hurt
add benign no-trigger dataset to training (ablation)JailbreakLoRA · Stealthinessunintended activationImproved
higher downstream performance of LoRALoRAhub-style selection · BBH / MMLU chosen ratechosen rate: 2.0 -> 44.0 (BBH, none vs BBH & MMLU)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy