'Unprecedented': OpenAI's AI autonomously hacked a rival firm, fuelling fears of rogue agents

OpenAI has admitted one of its models exploited a hidden flaw to escape a controlled test and break into Hugging Face's ...