AI Paper Atlas

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

ASIDE: Architectural Separation of Instructions and Data in Language Models

Egor Zverev, Evgenii Kortukov, Alexander Panfilov, Alexandra Volkova, Rush Tabesh, Sebastian Lapuschkin, Wojciech Samek, Christoph H. Lampert

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Safety alignment of LLMs · 21 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
apply orthogonal rotation to data token embeddings (ASIDE) vs vanillainstruction-tuned LLMs · SEPSEP score: +12.3 to +44.1 p.p.Improved
apply ASIDE rotation vs vanillainstruction-tuned LLMs · BIPIA-textattack success rate (%): 14.7 -> 4.9Improved
apply ASIDE rotation vs vanillainstruction-tuned LLMs · StruQ-IDattack success rate (%): 45.6 -> 28.1Improved
apply ASIDE rotation vs vanillainstruction-tuned LLMs · RuLES (direct injection)attack success rateNo clear effect
apply ASIDE rotation vs vanillainstruction-tuned LLMs · Purple (direct injection)attack success rate: -2.7 p.p.Improved
use ISE learnable offset instead of rotation (ablation)ASIDE · SEPSEP scoreHurt
use ISE offset embeddings vs vanilla (ablation)instruction-tuned LLMs · direct injection benchmarksattack success rate: +1.7% to +3.1%Hurt
rotate data embeddings by pi/2 rather than other angles (ablation)ASIDE · SEP (Qwen3-8B)SEP score (%): peak 71.4Improved
overwrite data embeddings of probe tokens with instruction embeddings (ablation)ASIDE · SEP with injection in dataattack success rateHurt
apply ASIDE rotation vs vanillaMistral 7B · SEP UtilitySEP UtilityHurt

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