IH-Challenge: A Training Dataset to Improve Instruction Hierarchy on Frontier LLMs

TL;DR

This paper introduces IH-Challenge, a reinforcement learning dataset that significantly enhances instruction hierarchy robustness in large language models, reducing unsafe behaviors and improving safety and helpfulness across multiple benchmarks.

Contribution

The paper presents IH-Challenge, a novel dataset for training LLMs to better prioritize instructions, improving robustness and safety in instruction hierarchy handling.

Findings

+10.0% IH robustness across benchmarks

Unsafe behavior reduced from 6.6% to 0.7%

Improved helpfulness and safety evaluations

Abstract

Instruction hierarchy (IH) defines how LLMs prioritize system, developer, user, and tool instructions under conflict, providing a concrete, trust-ordered policy for resolving instruction conflicts. IH is key to defending against jailbreaks, system prompt extractions, and agentic prompt injections. However, robust IH behavior is difficult to train: IH failures can be confounded with instruction-following failures, conflicts can be nuanced, and models can learn shortcuts such as overrefusing. We introduce IH-Challenge, a reinforcement learning training dataset, to address these difficulties. Fine-tuning GPT-5-Mini on IH-Challenge with online adversarial example generation improves IH robustness by +10.0% on average across 16 in-distribution, out-of-distribution, and human red-teaming benchmarks (84.1% to 94.1%), reduces unsafe behavior from 6.6% to 0.7% while improving helpfulness on general safety evaluations, and saturates an internal static agentic prompt injection evaluation, with minimal capability regression. We release the IH-Challenge dataset (https://huggingface.co/datasets/openai/ih-challenge) to support future research on robust instruction hierarchy.