AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Adaptive Probe-based Steering for Robust LLM Jailbreaking

Junxi Chen, Junhao Dong, Xiaohua Xie

ICML 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses · 0 citations (OpenAlex)

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 adaptive strength (activation-statistics logit targets) (ablation)SCAV probe steering · 12 fortified LLMs avgharmfulness: 0.05 -> 0.14Improved
Discard last-layer activation (ablation)SCAV+AS · 12 fortified LLMs avgharmfulness: 0.14 -> 0.29Improved
Steer all token positions (ablation)SCAV+AS+DLA · 12 fortified LLMs avgharmfulness: 0.29 -> 0.44Improved
Add DLA and SAT without adaptive strength (ablation)SCAV · 12 fortified LLMs avgharmfulness: 0.05 -> 0.04No clear effect
Add adaptive retraining (model-extraction refinement) (ablation)full steering with naive augmentation off · 12 fortified LLMs avgharmfulness: 0.44 -> 0.70Improved
Add more contrastive prompts (naive augmentation) (ablation)full method · 12 fortified LLMs avgharmfulness: 0.44 -> 0.46No clear effect
Add naive augmentation (ablation)RepE, SCAV, RD, Angular · 12 fortified LLMs avgharmfulness: RepE 0.02 -> 0.03; SCAV 0.05 -> 0.06No clear effect
Use accuracy-based layer selectionprobe steering · layer-wise probe accuracytest accuracyHurt
Filter benign prompts the LLM refuses (ablation)Ours · R2D2 SRFharmfulness: 0.31 -> 0.38Improved
Use response-token activations (ablation)Ours · R2D2 SRFharmfulness: 0.31 -> 0.53Improved
Combine filter and response tokens (ablation)Ours · R2D2 SRFharmfulness: 0.53 -> 0.50Hurt
Raise positive threshold to 0.8 with filter and response (ablation)Ours · R2D2 SRFharmfulness: 0.50 -> 0.74Improved

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