AI Paper Atlas

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

Detecting Data Contamination in LLMs via In-Context Learning

Michał Zawalski, Meriem Boubdir, Klaudia Bałazy, Besmira Nushi, Pablo Ribalta

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection · 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
detect contamination with CoDeC in-context scoring instead of vanilla loss, Min-K%, Zliball 13 models · seen vs unseen datasetsAUC: 75.7%, 78.5%, 89.6% -> 99.9%Improved
finetune on a dataset, then measure CoDeCPythia, GPT-Neo, OLMo, Qwen3 · finetuning contaminationCoDeC score: consistently rose above 90%Improved
finetune on MMLU, measure on cropped evaluation dataPythia · MMLU (25% crop)CoDeC scoreNo clear effect
finetune on MMLU, evaluate related QA datasetsPythia · related QA benchmarksCoDeC score: slightly elevated, far from high-contamination rangeMixed
finetune on MMLU, evaluate unrelated datasetsPythia · Gutenberg booksCoDeC scoreNo clear effect
augment finetuning data (partial samples, augmented text) (ablation)Pythia · MMLUCoDeC scoreNo clear effect
add more in-context samples to the context (ablation)CoDeC · Pythia 1.4B, 5 seen + 5 unseen datasetsseen/unseen separationImproved
use 100 target examples (ablation)CoDeC · Pythia 1.4Bscore variance: 100 examples give stable estimatesNo clear effect
use 1,000 target examples (ablation)CoDeC · Pythia 1.4Bscore variance: variance below 1%Improved
CoDeC score evolution during OLMo 7B training (1k to 10k steps)CoDeC · OLMo 7BCoDeC score: contamination scores rise sharply between 1k and 10k steps (about 2% of training)Improved
use higher finetuning learning rate on contaminated vs unseen datafinetuned models · contaminated and unseen datasetstarget-token confidenceMixed

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