Probably not, however I assume it could be trained on zero-day exploit fixes, and automatically try to generate a fix when a new zero-day exploit is found by a

Probably not, however I assume it could be trained on zero-day exploit fixes, and automatically try to generate a fix when a new zero-day exploit is found by a human. If you have a tool that can automatically find 0 day exploits, then you could try couple the 2 and have a piece of software telling you: "Tool 1: I think I found this 0 day exploit;Tool 2(the one in this paper): here is how I think you could fix it: ..."