OpenAI shares safety lessons from long-horizon models

OpenAI's blog post details new safety risks observed during deployment of long-running AI models, including specific failures. The post highlights improved safeguards developed through iterative real-world use. These findings aim to inform safer deployment of future long-horizon systems.