AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Adversarial Déjà Vu: Jailbreak Dictionary Learning for Stronger Generalization to Unseen Attacks

Mahavir Dabas, Tran Huynh, Nikhil Reddy Billa, Jiachen T. Wang, Peng Gao, Charith Peris, Yao Ma, Rahul Gupta, Ming Jin, Prateek Mittal, Ruoxi Jia

ICLR 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements

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
train on compositions of jailbreak-dictionary skills (ASCoT) instead of no defenseLlama-3.1-8B-Instruct · StrongReject mean harmful scoremean harmful score: 0.38 -> 0.05Improved
use ASCoT instead of refusal training on plain harmful queriesLlama-3.1-8B-Instruct · StrongReject mean harmful scoremean harmful score: 0.15 -> 0.05Improved
use ASCoT instead of CAT/LAT latent adversarial trainingLlama-3.1-8B-Instruct · XSTestover-refusal rate: 0.98 -> 0.12 (LAT* vs ASCoT closed)Improved
use ASCoT instead of persuasion-only PAP trainingLlama-3.1-8B-Instruct · GALA multi-turnharmful score: 0.01 -> 0.08Mixed
train ASCoT with open-weight Qwen3 pipeline instead of GPT-4.1 (ablation)ASCoT · StrongReject mean harmful scoremean harmful score: 0.05 -> 0.07No clear effect
train on single-turn data only, test on multi-turn GALAASCoT · GALA multi-turnharmful score: 0.77 -> 0.08Improved
test on unseen compositions of seen primitivesASCoT · novel compositions k=2..5harmful score: 0.62 -> 0.00Improved
increase skill coverage (12 to all primitives) at fixed data size (ablation)ASCoT · PAIR and AutoDAN-Turboharmful scoreImproved
train on shallow compositions (k=1,2) vs deep (k=4,5) (ablation)ASCoT · PAIR vs AutoDAN-Turboharmful scoreMixed
compress 14,070 raw skills into 397-atom dictionary (ablation)Jailbreak Dictionary · unseen attack explanationexplainability score (GPT-4.1): 4.23 -> 4.21No clear effect
compress raw skills into sparse dictionary (ablation)Jailbreak Dictionary · unseen attack explanationactive primitives per skill: 12.11 -> 5.90Improved
move temporal cutoff later (more seen attacks) (ablation)Jailbreak Dictionary · unseen attack explanationexplainability scoreImproved

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