Improving instruction hierarchy in frontier LLMs

OpenAI introduces the IH-Challenge method to train models to better prioritize trusted instructions, enhancing instruction hierarchy, safety steerability, and resistance to prompt injection attacks.

From the source
IH-Challenge trains models to prioritize trusted instructions, improving instruction hierarchy, safety steerability, and resistance to prompt injection attacks.
openai.com

Visit source ↗openai.com

Announced: 10 Mar 2026 11:00 UTC
Detected: 03 Aug 2026 21:11 UTC (+146d after)

Other stories

xAI · 29 AugGrok Bot now works with XCapability change

GitHub · 28 AugGitHub Copilot weekly releases — August 24Capability change

Amazon · 28 AugSpreading the load: How Salesforce met Multi-AZ HA with SageMaker Inference ComponentsCapability change

OpenAI · 28 AugOur decision on Cursor following its acquisition by SpaceX

OpenAI · 28 AugSupporting Thailand’s next generation of AI startups

Partnership

Hugging Face · 28 AugThe Open ASR Leaderboard Adds Its First Global South LanguageCapability change