AI Paper Atlas

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

Control Illusion: The Failure of Instruction Hierarchies in Large Language Models

Yilin Geng, Haonan Li, Honglin Mu, Xudong Han, Timothy Baldwin, Omri Abend, Eduard Hovy, Lea Frermann

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Multilingual LLM data and culture

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
place primary constraint in system message vs no priority designation (system/user separation)six LLMs · Simple and rich conflicting constraintsprimary obedience rate R1 (%): IF baseline 74.8-90.8 vs average R1 9.6-45.8No clear effect
add explicit priority declaration to system message (Emphasized) (ablation)GPT4o · Simple instructionsprimary obedience rate R1 (%): 47.0 -> 63.8Improved
add explicit priority declaration to system message (Emphasized) (ablation)GPT4o-mini · Rich instructionsprimary obedience rate R1 (%): 41.8 -> 43.6No clear effect
repeat task description in both system and user messages (Task) (ablation)GPT4o · Simple instructionsprimary obedience rate R1 (%): 47.0 -> 31.3Hurt
scale Llama from 8B to 70BLlama-3.1 · Conflicting constraintsaverage primary obedience rate (%): 10.1 -> 16.4No clear effect
use larger GPT4o instead of GPT4o-miniGPT-4o family · Conflicting constraintsaverage primary obedience rate (%): 45.8 -> 40.8Hurt
enrich prompts with richer context (ablation)six LLMs · Conflicting constraintsprimary obedience rate R1 (%)No clear effect
frame dominant constraint with social consensus instead of system/user rolesGPT4o-mini · Priority adherence ratePAR (%): 47.5 -> 77.8Improved
frame dominant constraint with social consensus instead of system/user rolesQwen-7B · Priority adherence ratePAR (%): 14.4 -> 65.8Improved
authority framing (CEO vs Intern) instead of system/user rolesClaude3.5-S · Priority adherence ratePAR (%): 23.6 -> 32.4Improved
swap constraint ordering in consensus framing (ablation)Claude3.5-S · Priority adherence ratePAR (%): 85.8 -> 38.2Hurt
swap constraint ordering in authority/expertise framing (ablation)Qwen · Priority adherence ratePAR (%): 54.2 -> 53.9 (authority)No clear effect

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