Improving instruction hierarchy in frontier LLMs 文章

OpenAI Blog2026-03-10BLOGen

摘要

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

相关事件

暂无数据

相关公司

暂无数据

相关人物

暂无数据

相关产品

暂无数据