Even the most basic of AI tool, assuming they used them for their terraforming, would know better.

You would think so, but no. Not in my experience, at least. Especially if you're focused on a narrow goal or fixing a very specific problem, AI agents tend to fixate on the happy path to the neglect of everything else. They may start out seeing the big picture, but it takes almost no time at all for stuff like security to drop from their attention unless you're actively reminding them.

This is exactly the kind of issue I've seen AI cause at the enterprise I work for.