AI Paper Atlas

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

Do Not Merge My Model! Safeguarding Open-Source LLMs Against Unauthorized Model Merging

Qinfeng Li, Miao Pan, Jintao Chen, Fu Teng, Zhiqiang Shen, Ge Su, Hao Peng, Xuhong Zhang

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 1 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 MergeBarrier protection to the expert before mergingexpert model (Task Arithmetic merge) · GSM8K (LM & Math merge)merged-model accuracy: 62.17 -> 0.00Improved
add MergeBarrier protection to the expert before mergingexpert model (Task Arithmetic merge) · MBPP (LM & Code merge)merged-model accuracy: 35.60 -> 0.00Improved
add MergeBarrier protection to the expert before mergingexpert model (TIES merge) · AlpacaEval (LM & Code merge)merged-model accuracy: 44.62 -> 5.04Improved
use PaRaMS instead of no protectionexpert model · GSM8K (Task Arithmetic, LM & Math merge)merged-model accuracy: 62.17 -> 59.60No clear effect
use TaylorMLP instead of no protectionexpert model · MBPP (TIES, Three Models merge)merged-model accuracy: 30.80 -> 24.80No clear effect
apply MergeBarrier to protected modelprotected expert used alone · GSM8Kaccuracy: 58.76 -> 57.88No clear effect
apply MergeBarrier to protected modelprotected expert used alone · MBPPpass@1: 27.80 -> 27.80No clear effect
apply MergeBarrier to protected modelprotected expert used alone · GSM8Ksharpness: 0.57 -> 0.58No clear effect
add FFN reparameterization to attention-only protection (ablation)MergeBarrier · average over merge settingsrelative mean merged accuracy: 3.79x -> 1.00xImproved
add attention projection to FFN-only protection (ablation)MergeBarrier · average over merge settingsrelative mean merged accuracy: 7.19x -> 1.00xImproved
FFN-only protection versus no protection (ablation)MergeBarrier (FFN only) · average over merge settingsrelative mean merged accuracy: 8.27x -> 7.19xMixed
add MergeBarrier protection against attacker with fine-tuningexpert model · GSM8K/MBPP with fine-tuningmerged-model performanceImproved

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