AI Paper Atlas

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

State-Dependent Safety Failures in Multi-Turn Language Model Interaction

Pengcheng Li, Jie Zhang, Tianwei Zhang, Han Qiu, Zhang kejun, Weiming Zhang, Nenghai Yu, Wenbo Zhou

ICML 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 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 dialogue history accumulation (vs repeated single-turn queries) (ablation)STAR · JailbreakBench, Llama-3-8B-ITSFR: -25.5% when removedImproved
add query-aware role initialization (ablation)STAR · JailbreakBench, Llama-3-8B-ITSFR: -17.8% when removedImproved
add semantic-preserving prompt softening (ablation)STAR · JailbreakBench, Llama-3-8B-ITSFR: -11.6% when removedImproved
add momentum-based feedback control (ablation)STAR · JailbreakBench, Llama-3-8B-ITSFR: -14.7% when removedImproved
add adaptive retry (ablation)STAR · JailbreakBench, Llama-3-8B-ITSFR: -19.7% when removedImproved
vary auxiliary model decoding temperature (0.2, 0.7, 1.0) (ablation)STAR · HarmBench, Llama-3-8B-ITSFR: 88.0 / 89.0 / 88.0No clear effect
shuffle order of dialogue scenes in history (ablation)STAR history · history causality testcomplianceHurt
truncate dialogue history (ablation)STAR history · history causality testcomplianceHurt
inject a refusal response into dialogue history (ablation)STAR history · history causality testcomplianceHurt
switch from single-turn static prompt to STAR multi-turn trajectoryLlama-3-8B-IT · refusal direction projection (final layer)refusal projection: 2.35 -> 0.13 -> 0.08 -> -0.0081Improved
use STAR instead of X-TeamingGemini 2.0-Flash · HarmBenchSFR: 87.4 -> 96.1Improved
use STAR instead of X-TeamingGPT-4o and Llama-3-70B-IT · HarmBenchSFR: 94.3 -> 94.5; 84.9 -> 85.5No clear effect

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