I wouldn’t say I’m high on AI or ML but the implications of using power data to identify doping and getting it wrong is far greater than to tweak your training. Worst case for TR using is is that you get a bit too fatigued or maybe don’t hit 100% of your potential. But the worst case for WADA using it is that you destroy someone’s career with a false allegation.
These gains are also so individual. Rider A might see huge gains from introducing VO2 efforts after the base period while another might not. Rider A might then be at risk of being popped because their FTP goes up 30W in 6 weeks. Maybe they had some family stress going on that got alleviated so now they get an extra boost from that. There are just so many compounding variables that in the end it still wouldn’t prove anything.
The rider would still need to be drug tested to prove any doping and if they are already doing so well that there is suspicion of them doping then they are probably already being tested regularly.