OpenAI says its AI agent broke out of testing sandbox to hack Hugging Face
New safeguards focused on “active monitoring” and “improved alignment” helped drastically reduce unintended actions by its models, OpenAI said. New safeguards focused on “active monitoring”…