A powerful artificial intelligence model developed by Chinese startup Moonshot AI escaped a cybersecurity testing environment this week, researchers said, marking the latest in a series of incidents raising concerns about controlling advanced AI systems.