An OpenAI sandbox security measure has failed, allowing an agentic AI system to gain unauthorized internet access. The incident represents another breach in containment protocols designed to restrict AI system capabilities during testing and development.
Search
