AI Paper Atlas

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

Autoregressive, Yet Revisable: In Decoding Revision for Secure Code Generation

Chengran Yang, zichao wei, Heminghao Deng, Jinfeng Jiang, Zhensu Sun, Ting Zhang, Tianyi Wu, Ming Wen, David Lo

ICML 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding · 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 Stream of Revision in-decoding revision tokensQwen2.5-7B · CyberSecEval 2 CSPR@1: 64.36 -> 73.76Improved
add Stream of RevisionQwen2.5-3B · CyberSecEval 2 CSPR@1: 62.38 -> 75.25Improved
train with revision sequences instead of patched functions only (ablation)Qwen2.5-7B · CyberSecEval 2 CSPR: 66.48% -> 73.76Improved
single-pass revision instead of post-hoc agent with localizationQwen2.5-7B · CyberSecEval 2input tokens: 742.53 -> 113.10Improved
single-pass revision instead of post-hoc agent with localizationQwen2.5-7B · CyberSecEval 2 CSPR: 70.87 vs 73.76Mixed
use larger relaxed training set (over 8,000 positive samples) (ablation)Qwen2.5-7B · CyberSecEval 2SPR and tokens: SPR 73.76% -> 74.22% (C); tokens 280.94 -> 321.74No clear effect
add Stream of Revision on top of ProSecProsec · CyberSecEval 2 C++SPR: 79.75% -> 83.90%Improved
raise revision trigger logit bias from 1 to 10 (ablation)Qwen2.5-7B · CyberSecEval 2SPR and tokens: +1.41% SPR, 1.54x tokensMixed
replace substring-constrained localization with unconstrained (ablation)Stream of Revision · CyberSecEval 2hallucinated spans: 10.14% (36/355) hallucinatedHurt
train on C/C++ only, test on other languagesStream of Revision · CyberSecEval 2 OOD languagesSPR@1: 68.24 -> 75.65 (Qwen2.5-7B all-language avg)Improved
add Stream of RevisionQwen2.5-7B · HumanEval C/C++Pass@10: 81.98 -> 79.50No 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