If an AI can tell what’s in a picture (which is a series of pixel data), it can certainly reconstruct gcode into an “image.” Ive used it at work for similar tasks.

Cylinders obviously are too generic, but there are parts that are a lot more specific and could be fingerprinted.