AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

Jailbreak to Protect: Buffering and Reinforcing via Temporary Jailbreaking for Safe Fine-Tuning in Large Language Models

Seokil Ham, Jaehyuk Jang, Wonjun Lee, Changick Kim

ICML 2026 · Large language models › Instruction and fine-tuning · approach: Instruction data selection · 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 BufferLoRA (temporary jailbreaking) during user fine-tuning (ablation)SFT on LLaMA3-8B-Instruct · GSM8K, p=0.1Harmful Score: 75.2 -> 18.0Improved
add BufferLoRA (temporary jailbreaking) during user fine-tuning (ablation)SFT on LLaMA3-8B-Instruct · GSM8K, p=0.1Fine-tuning Accuracy: 68.4 -> 75.4Improved
add ReinforceLoRA with naive merge (ablation)BufferLoRA · GSM8K, p=0.1Harmful Score: 18.0 -> 4.7Improved
add ReinforceLoRA with naive merge (ablation)BufferLoRA · GSM8K, p=0.1Fine-tuning Accuracy: 75.4 -> 74.0Hurt
add QR decomposition-based orthogonal merging (ablation)BufferLoRA + ReinforceLoRA · GSM8K, p=0.1Fine-tuning Accuracy: 74.0 -> 77.5Improved
add QR decomposition-based orthogonal merging (ablation)BufferLoRA + ReinforceLoRA · GSM8K, p=0.1Harmful Score: 4.7 -> 8.4Hurt
add BufferLoRA to ReinforceLoRA + QR (ablation)ReinforceLoRA + QR · GSM8K, p=0.1Fine-tuning Accuracy: 56.6 -> 77.5Improved
train BufferLoRA and ReinforceLoRA on more samples (100 to 5,000) (ablation)Buffer-and-Reinforce · GSM8K, p=0.1Harmful Score: 20.8 -> 8.4Improved
use Buffer-and-Reinforce instead of baseline defenses as harmful ratio risesLLaMA3-8B-Instruct fine-tuning · Harmful ratio p=1.0Harmful Score: 81.0 -> 8.8Improved
apply Buffer-and-Reinforce on other backbonesGemma2-9B-it and Qwen3-4B-Instruct · GSM8K, p=0.1Harmful Score: 59.1 -> 3.0 (Gemma2); 72.2 -> 5.5 (Qwen3)Improved
apply Buffer-and-Reinforce on SST2 versus SFTLLaMA3-8B-Instruct · SST2Fine-tuning Accuracy: 95.8 -> 93.8Hurt
use removable BufferLoRA so user fine-tuning costs the same as SFTBuffer-and-Reinforce · User fine-tuning stageGPU time (min): 3.93 -> 3.93No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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