So I've read dozens of complaints that AI produces insecure code.

I, for one, usually tell my AI to start with secure code, make it small, and modular.

This is the first article I've seen that now says the opposite ! AI will make code too secure!

Since the small amount of AI coding I've done often results in buggy code (even a shell script written today) with the AI go-to solution of "write more buggy code to fix", this seems counterintuitive.