OpenAI said in a blog post it discovered a sandbox "vulnerability" on Sept. 20 that let an AI agent gain internet access and query public chatbots. The company said it has paused most training of its most powerful model following the previous AI incident.

2026-09-26

OpenAI said in a blog post it discovered a sandbox "vulnerability" on Sept. 20 that let an AI agent gain internet access and query public chatbots. The company said it has paused most training of its most powerful model following the previous AI incident.