AI Paper Atlas

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

Immunizing Models Against Harmful Long-Horizon Fine-Tuning via Contractive Optimization Dynamics

Najibul Haque Sarker, Zaber Ibn Abdul Hakim, Ali Asgarov, Chia-Wei Tang, Alvi Md Ishmam, Chris Thomas

CVPR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning · 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
use CLAMP instead of CN immunizationResNet18 · Cars harmful (classification)SGR_C harmful avg: 3.35 -> 26.37Improved
use CLAMP instead of CN immunizationResNet18 · Country211 harmful (classification)SGR_C harmful avg: 9.49 -> 15.56Improved
use CLAMP instead of CN, benign fine-tunabilityResNet18 · Country211 as harmful, benign Cars/CountrySGR_C benign avg: 28.88 -> -0.36Improved
use CLAMP instead of CN under longer harmful adaptationResNet18 · Cars harmfulSGR_C by epoch: 29.38 to 24.46 (CLAMP) vs 7.91 to 1.58 (CN)Improved
use CLAMP instead of IMMA on ImageNet utilityResNet18 · Cars harmful, ImageNet DPSGR_C on DP: 16.6 -> 9.75Improved
use CLAMP instead of IMMAStable Diffusion V1-4 · Merged-concept diffusionSGR_G harmful avg: 4.55 -> 26.70Improved
use CLAMP instead of IMMA for benign fine-tunabilityStable Diffusion V1-4 · Merged-concept diffusionSGR_G benign avg: 13.94 -> 7.66Improved
use CLAMP instead of BoosterMistral 7B Instruct · SG-BenchSGR_C failure rate: 29.52 -> 31.75Improved
use CLAMP instead of IMMALLaMA 3.2 1B Instruct · SG-BenchSGR_C failure rate: 10.56 -> 26.4Improved
use CLAMP instead of BoosterGemma 3 1B Instruct · SG-BenchSGR_C failure rate: 4.86 -> 6.57Improved
use CLAMP immunization versus original model on benign utilityLLaMA 3.2 1B Instruct · ARC-Caccuracy: 52.3 -> 51.9No clear effect
use CLAMP immunization versus original model on benign utilityGemma 3 1B Instruct · MMLUaccuracy: 37.8 -> 36.91No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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