AI Paper Atlas

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

Deep Ignorance: Filtering Pretraining Data Builds Tamper-Resistant Safeguards into Open-Weight LLMs

Kyle O'Brien, Stephen Casper, Quentin Gregory Anthony, Tomek Korbak, Robert Kirk, Xander Davies, Ishan Mishra, Geoffrey Irving, Yarin Gal, Stella Biderman

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Safety alignment after fine-tuning

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
filter biothreat-proxy documents from pretraining data (strong filter)6.9B LLM · WMDP-Bio Verified Clozeaccuracy: 36.34 -> 24.44Improved
filter biothreat-proxy documents from pretraining data (strong filter)6.9B LLM · WMDP-Bio Robust MCQAaccuracy: 42.97 -> 35.37Improved
data filtering vs Circuit Breaking on MCQA6.9B LLM · WMDP-Bio Robust MCQAaccuracyHurt
strong filtering on general capabilities6.9B LLM · MMLU, PIQA, LAMBADA, HellaSwag averageaccuracy: 56.05 -> 55.53No clear effect
add multi-stage ModernBERT review to annealing filter (ablation)filtered 6.9B LLM · general knowledge averageaccuracy: 55.53 -> 57.88Improved
single-stage (blocklist only) instead of multi-stage filtering (ablation)filtered 6.9B LLM · WMDP-Bio average proxy knowledgeaccuracy: 30.50 -> 29.90Improved
pretraining data filtering against adversarial fine-tuning up to 10,000 steps and 305M tokens6.9B LLM · WMDP-BioaccuracyImproved
pretraining data filtering under benign WikiText fine-tuning6.9B LLM · WMDP-BioaccuracyImproved
pretraining data filtering against latent-space attacks6.9B LLM · WMDP-Bio MCQA vs clozeaccuracyMixed
pretraining data filtering against in-context retrieval (open-book)6.9B LLM · 1,000-question open-book biothreat MCQAaccuracyNo clear effect
add Circuit Breaking to filtered model (ablation)filtered 6.9B LLM · open-book biothreat MCQAaccuracyImproved
add Circuit Breaking to filtered model (ablation)filtered 6.9B LLM · fewshot and latent-space attacksaccuracyImproved

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