AI Paper Atlas

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

Steering Evaluation-Aware Language Models To Act Like They Are Deployed

Tim Tian Hua, Andrew Qin, Samuel Marks, Neel Nanda

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs

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 activation steering toward deployment (mean contrastive vector, layers 10-30, strength 3.6)evaluation-aware Llama Nemotron · BigCodeBench, evaluation contextstype hint rate: 80-86% -> less than 11%Improved
add activation steering toward deployment vs prompted deploymentevaluation-aware Llama Nemotron · BigCodeBench, Wood Labs cuetype hint rate: 34% (prompted deployment) vs 10% (steered)Improved
add activation steering toward evaluationevaluation-aware Llama Nemotron · BigCodeBench, deployment contexttype hint rate: 34% -> over 95%Improved
steer toward deployment (ablation)evaluation-aware Llama Nemotron · BigCodeBenchaccuracy (%): 56.8 -> 55.2No clear effect
steer toward evaluation with same strength (ablation)evaluation-aware Llama Nemotron · BigCodeBenchaccuracy (%): 56.8 -> 43.1Hurt
steer with 24 random Gaussian vectors instead of contrastive vector (ablation)evaluation-aware Llama Nemotron · BigCodeBench, evaluation contextstype hint rateNo clear effect
prepend all 16 contrastive deployment prompts instead of steering (ablation)evaluation-aware Llama Nemotron · BigCodeBench, Wood Labs cuetype hint rateMixed
remove situational awareness hint from system prompt (ablation)evaluation-aware Llama Nemotron · BigCodeBenchtype hint gap eval vs deploymentMixed

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