Security debt ahoy: Only about half of the code that the latest large language models (LLMs) create is cybersecure, and more and more of it is being created all the time.