Same. I write cheats for CS2 for my own amusement and it's 100% detectable by replay analysis.

If I'm reacting to information that I shouldn't have at a rate much much greater than the general population, then I'm not psychic, I'm just cheating.

This is also true for aiming, spray control, etc. All current cs2 cheaters are producing a very large and detectable audit trail of very suspicious plays in game, even if they think they're being sneaky. The resolution of the data looks like keeping track of your location and where you're aiming about every 15ms.

Here's some recent research that's related: https://arxiv.org/html/2508.06348v1

I can hide from a kernel module, but I can't hide from my own data trail.

There's only really two paths:

* We do kernel anti-cheat in trusted execution environments, which is bad for all the reasons in the article. When I break through this, you'll get the full "having a cheater" experience in your game.

* We do AI/ML heuristic-based detection to the point where cheaters are forced to behave exactly like non-cheating teammates or risk detection, cheating maybe only 10-20% above their previously established skill patterns. When I break through this, you'll have a normal game and I'll be kind of bored and nobody will be having the "cheaters in my game" experience even though I'm actually cheating.

In either case, I'm still going to try and beat the system for fun. Because video games.

Heuristic detection is iffy IMO, because while this might look trivial now (=> before cheaters optimize for it) false positives could easily ruin the user experience after a few rounds of cat & mouse: Even if you manage to keep sensitivity high (detecting >90% of cheaters), if you have like single digit percentages of cheaters and false positives, a lot of your bans are innocent.

If you keep the heuristics "tolerant", on the other had, you end up perpetually chasing in a cat-and-mouse game, and a lot of cheaters go undetected inbetween detection upgrades.

I feel heuristic detection is a "metagame" that developers don't really want to engage in, because you basically can't really win and it's perpetual work/spending.

Valve has already discussed using ai? There's also some commercial alternatives. I haven't seen that it works well yet, so it doesn't seem easy to solve.

But it seems that will just push cheat devs to adjust their aimbot to become more human in behavior? For the average gamer the top 0.1% will look like cheaters anyway.