That's just very good programming at work, if you ask any simulation developer about their AI guaranteed you'll get the AI runs a lower precision physics simulation than the player car and we program it to react like a human, i.e. program in limitations, mistakes, bad judgement, losing concentration. I highly doubt back then that AI was as complex as it is today, I mean Google spent years crafting an AI to react human like...I doubt some game developers stumbled upon human like AI for a game back then(the subroutines would be larger than the entire game itself). There is no chance the AI wasn't cheating then, AI is too vast and complex for it to be robust enough back then to not resort to cheating like having reduced physics calculations, instantaneous ability to deal with loss of grip, knowing braking distances on any type of surface and any tire condition...lets be real all AI cheat. Good programming can make it hard to tell, but they are all cheating.