AI Paper Atlas

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

Steering Beyond the Support: Adversarial Training on Unsupervised Jailbroken Activation Simulation

LUOYU CHEN, Weiqi Wang, Zhiyi Tian, Chenhan Zhang, Feng Wu, Jianhuan Huang, Ahmed Asiri, Shui Yu

ICML 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

Read the paper

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
Use unsupervised latent-direction adversarial training instead of targeted-prefix adversarial training (ablation)Potential-field steering · HarmBench, LLaMA-3-8BAvg SR / coverageImproved
Add the adversarial inner loop (AT) to bi-level training (ablation)Potential-field steering · HarmBench, LLaMA-3-8BAvg SR / coverageImproved
Train steering on simulated jailbroken activations without jailbreak supervision (vs supervised AlphaSteer)Safety steering · HarmBench, Mistral-v2-7BStrongReject (%): 5.86 -> 5.26Improved
Use the proposed steering instead of CB for over-refusalLLaMA-3-8B · OR-Bench boundaryFPR (%): 49.0 -> 16.6Improved
Supervised steering trained on a subset of jailbreak families, tested on unseen familiesAlphaSteer · HarmBench, Mistral-v2-7BStrongReject (%)Hurt
Attacker adapts to the defense (adaptive GCG)Ours · HarmBench GCG, Mistral-v2-7BStrongReject (%): 7.10 -> 12.55Hurt
Steering-aware adaptive GCG minimizing steering-gradient normOurs · HarmBench GCG, Mistral-v2-7BStrongReject (%): 7.10 -> 15.54Hurt
Increase number of latent directions K from 128 to 512 (ablation)ULDD simulation · LLaMA-3-8B# successful simulated jailbreaks: 62 -> 148Improved

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