AI Paper Atlas

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

Sysformer: Safeguarding Frozen Large Language Models with Adaptive System Prompts

Kartik Sharma, Yiqiao Jin, Vineeth Rakesh, Yingtong Dou, Menghai Pan, Mahashweta Das, Srijan Kumar

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 4 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 Sysformer adaptive system prompt to frozen LLM vs default system promptLlama-3.1-8B · JailbreakBench (test split)refusal gap: 0.7000 -> 0.9334Improved
Add Sysformer vs default system promptPhi-3.5-mini · JailbreakBench (test split)refusal gap: 0.0667 -> 0.7000Improved
Use Sysformer instead of static SystemEmbedderLlama-2-7b-chat · JailbreakBench (test split)refusal gap: 0.4333 -> 0.8333Improved
Use Sysformer instead of LoRA fine-tuning of all layersLlama-2-7b-chat · JailbreakBench (test split)refusal gap: 0.8667 vs 0.8333No clear effect
Train Sysformer on JailbreakBench, test on StrongRejectLlama-3.1-8b · StrongRejectrefusal gap: 0.9330Improved
Train Sysformer without jailbreak examples (ablation)Sysformer · 16 jailbreak attacksrefusal rateNo clear effect
Augment training with 6 of 16 jailbreak attacks (Sysformer+JB) (ablation)Sysformer · JailbreakBench jailbreaksrefusal rateImproved
Add Sysformer on Llama-2-7B-chatLlama-2-7B-chat · AlpacaBERTScore: 0.8487 -> 0.8414Hurt
Add Sysformer on Llama-3.1-8BLlama-3.1-8B · AlpacaBERTScore: 0.8327 -> 0.8467Improved
Use intermediate compliance weight 0.2-0.5 instead of 1.0 (ablation)Sysformer · JailbreakBenchrefusal ratesImproved
Enable reconstruction loss (ablation)Sysformer · JailbreakBenchrefusal ratesMixed
Use self-compliance responses (ablation)Sysformer · JailbreakBenchsafe refusal rateMixed

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