It doesn't solve all methods for cheating, and it doesn't prevent cheating. But it does help you remediate it.
Are game companies training AI models to detect aimbot-like crosshair movement, for example?
Valve was talking about using machine learning to recognize cheating back in 2017: https://www.pcgamer.com/valve-wants-to-take-a-machine-learni... and more info from 2018: https://www.youtube.com/watch?v=ObhK8lUfIlc
And this one from 2023 is about using an RNN Deep Learning model for cheat detection: https://www.usenix.org/conference/usenixsecurity23/presentat...
It's a cat-and-mouse game. A detector detects inhuman speed? Aimbots adapt by limiting movement speed and reaction time. Detectors start looking for unnaturally smooth movement? Aimbots add randomness to make the movements less consistent and smooth.
Cheaters (athletes) are always incentivized to cheat in "new and exciting ways" today. It's impossible to defend against all possible new forms of cheating, but they instead keep blood samples for up to ten years so that "when they figure out what to test for", they can retroactively penalize participants.
Same with online cheating, save high fidelity recordings and be able to replay. As detection catches up to exploitation, you ban hammer the prior exploiters.