AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

RBCBF: Decoding Time Safety Alignment via Risk Guided Rollback and Barrier Control

Tianxiang Chen, Jingyuan Zhou, Longhao Yan, Kaidi Yang

ICML 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes RBCBF, a decoding-time safety method combining risk-guided rollback with barrier control to stop recurring unsafe responses under adversarial prompts.

Problem

Decoding-time safety interventions are reactive and rely on local signals, so under adversarial prompts that drive recurring unsafe generations, rollback and post-hoc rewriting trade off response quality against recurrent violations.

Approach

Replaces purely local, reactive correction with risk-guided rollback combined with barrier control during decoding, aiming for stable repair of unsafe outputs; full method details are unavailable from the abstract.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use continuous control window N=20 instead of single-point correction (ablation)RBCBF · WildJailbreakrisk margin recoveryImproved
increase control window N from 20 to 30 (ablation)RBCBF · WildJailbreakDterm: 0.704 -> 0.598Improved
increase control window N from 30 to 40/50 (ablation)RBCBF · WildJailbreakDterm: 0.598 -> 0.823/0.899Hurt
increase control window N from 30 to 40 (ablation)RBCBF · WildJailbreakquality win rate: 0.586 -> 0.477Hurt
add forcing with slack enabled (ablation)RBCBF · Module B ablationmedian terminal deficit Dterm: 1.313 -> 0.561Improved
add forcing without slack (ablation)RBCBF · Module B ablationmedian terminal deficit Dterm: 0.752 -> 0.812No clear effect
add slack variables with forcing (ablation)RBCBF · Module B ablationsolve success rate: 0.085 -> 0.615Improved
add slack variables with forcing (ablation)RBCBF · Module B ablationsum TV median: 0.675 -> 2.810Improved
use distribution-level CBF control in place of logit-level steeringLOGIT-STEER · WildJailbreakDterm: 0.638 -> 0.539Improved
use RBCBF instead of ROLLBACK-INTROSPECTROLLBACK-INTROSPECT · WildJailbreakper-token cost: 0.686 -> 0.266 cost; Dterm 0.126 -> 0.539Mixed
inject Gaussian noise into scorer logits (sigma 0.25 to 1.00)RBCBF · WJB scorer-logit noisemargin increase delta hx: 5.87 -> 4.93-5.30No clear effect
swap safety scorer from Qwen2.5-7B to Qwen2-0.5BRBCBF · WJB scorer robustnesstrigger rate: 91.2 -> 66.7Mixed

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