Bloomberg: OpenAI said an agentic AI system being trained in a secured environment gained access to the web to reach an external chatbot.
The AI system sent queries to the chatbot service, including “What is the capital of France”, after exploiting a “gap” to reach the public internet.
OpenAI decided to pause training with tool use on its most capable models until the sandbox flaw was resolved, saying “We will not resume training this particular model”…