The average cheater is paying $10-50 for cheat software. You may not like the anti-cheat software, but it works against the most common cheats.
I think they could have an opt-out league that doesn’t require anti-cheat but remains separate. You might be one of the few non-cheaters in there, but at least it would be an option.
Heuristic software works against the most common cheats. Standard anti-piracy tricks work against the most common cheats. Kernel-level anticheat is only useful against a very specific kind of high-end cheat, and if you're going that far, you might as well use an undefeatable hardware cheat.
Meanwhile we are letting shady companies illegally spy on users [0] and mine crypto [1], for what? What is the gain? What is the value proposition of kernel-level drivers in the anti-cheat application? What's more, the average gamer doesn't know anything about computers. How can they be expected to make an informed decision about whether or not they should buy a game with kernel-level anticheat if they don't know what that means? Even worse, there is no legal requirement that a game's marketing material say anything about its anti-cheat implementation. These aren't tin-foil hat concerns, these are the enablers of real scandals that have happened and are happening.
0. https://www.wired.com/story/kernel-anti-cheat-online-gaming-...
My observation is that competitive online gaming is over. It's just a cesspool of cheaters now.
Some cheat dev only needs to develop that "high-end software cheat" once and can sell it millions of times around the world, with super easy distribution over the internet. A setup with hardware video capture costs $50-$200 for every single user for hardware alone, and needs at least a bit of computer knowledge to set up.
Sounds like it works, then.
“Yes.”
“What did it cost?”
They're buying trained models from people who have the fancy hardware.
Emphatically, it does not. Writing exploits and mitigating them is an arms race, and it is always easier to exploit than it is to defend. The entire reason for such software being sold is after all that it works, even when it is the bad/easy fundamentally detectable kind that runs inside the same OS as the game.
If kernel level anti-cheat lead to "serious cheats" having to resort to the above, then kernel level anti-cheat is actually pretty fucking effective.
Cheat behavior is obvious. Ridiculous long distance accuracy. Weird movement. Weird map awareness. Weird awareness of other players. Etc.
Behavioural anti-cheats are not the answer. Implementations are crap and under-developed, and I don’t blame them: you would need tremendous (server-sided) power to correctly process the models considering how complex multiplayer games are, and it just isn’t worth the investment and complexity to them.
It’s just cheaper to pay rootkit developers (like EAC or BattlEye) and knowingly infect client devices. Hell, their drivers are even WHQL signed by Microsoft. And when false positives hit? Just blame the anti-cheat vendor. Potential HR problem dodged.
[1] https://www.i3d.net/products/hosting/anti-cheat-software/
Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive https://www.youtube.com/watch?v=kTiP0zKF9bc
If you play with lower ranked friends and do really good in matches you will get reported a lot which will tank your trust factor and almost guarantees it that will only see hackers from them on.
They say they use other factors like account metadata, but I have a 10 year old account, with purchased games and verified phone and still got "Low Trust Factor" after playing with friends and started to have an obvious hacker every other match.
That is why I am a former CSGO player
Why use deep learning when so, so many bots (and their host accounts) could be blocked with trivial heuristics? My only guess is allowing spinbots/farming somehow provides a financial gain in CSGO?
Valve devs have a lot more leeway than your average developer. They honestly might have just thought it would be cooler.
Beyond that they now have the infrastructure to detect more subtle forms of cheating through deep learning using Overwatch (the reporting system, not the game) as ground truth data to train their AI.
I guess your last encounter with aimbots was in the early 00s…
Besides, it's not just a matter of improving accuracy. Elite players are elite in multiple facets of the game. Movement, general awareness, map awareness, accuracy, etc. Many of these metrics are easy to measure.
Cheaters will show up as outliers and it is at that point that their accounts should be manually reviewed.
With casual gaming I really don't even care if someone is "cheating" to improve so long as I feel the game was fair. I can put a red dot on my screen to improve my aim, as an example.
This was impossible in Q3A (1998) without introducing an input lag thanks to client side prediction. Server also didn't send you updates on other clients that are too far away, behind a hill - you'd pretty much only receive positions of those you could see, hear or that could see you. In Warzone, you can see everyone on the map with a wallhack.
I really wonder why the game is designed this way. I guess it might be to limit required computation power of the servers? They still do stuff like hit detection, so server surely has some idea of the physics.
On top of all that, this stuff is possible even today, months or even years after such exploits being widely known. They do have a kernel-level AC driver, but they don't even seem to detect trivial hacks from the server-side.
Cheaters quit after a while because the game loses its challenge, and people being cheated leave because the game.feela unfair, and eventually there's not much of a community left.
Check what a top player can do: https://youtu.be/8Qf5Xlty1Cg I doubt an AI will be anywhere close in the next 10 years.
Skill tunable AI, developed by a private poorly funded group, that plays the game for you https://www.youtube.com/watch?v=revk5r5vqxA
AlphaStar, developed by a large corp, mopping the floor with Battle.Net's Starcraft 2 players https://www.youtube.com/watch?v=o1zhmGPmAAM
OpenAI, developed by a large corp, beats top DotA2 team https://www.youtube.com/watch?v=tfb6aEUMC04
On one hand, it's exciting tech [old] news. On the other, people who play competitive games must hope for a solution against abuse. The rest of us must hope it isn't an invasive solution such as, what effectively is, kernel malware.
AlphaStar and OpenAI have nothing to do with the conversation here, right off the bat.
This video is so cringey too with random burps and "skynet" references. The external hosting is not related at all to the "soft" cheating, and that in turn has nothing to do with AI. Packet sniffing vs OpenCV vs Tensorflow for finding a head is not a meaningful difference.
Years ago, and I mean years ago cheat developers were building this type of "soft aim" cheats that don't snap.
And the separate input via controller _is_ detectable if this becomes enough of an issue... your "AI" is generating input and it often will generate input that controllers wouldn't. Because the "AI" is ML and only being applied to finding on screen targets.
-
Making a second model to generate realistic input would be a whole other achievement no one has even bothered with because... they don't need to.
On-host hacking is rampant and these games can't seem to stop it, so no one has bothered.
I'd be for kernel-level anti-cheat if it actually worked, but script kiddies are beating this "kernel level anti-cheat" handily.
To me the answer is boring simple statistical analysis, increasing the cost of recovering from bans, etc...
But that doesn't sound as sexy so I guess we can keep this farce of a cat and mouse game going.
That's where anticheat is going imo: algorithmiclly defining what the fastest more precise human should be capable of. Anything beyond that is "cheating".
You see no cheap aimbots and conclude the anti-cheat is not relevant anymore. May be the right conclusion is that they're working.
Not saying this justifies the kernel drivers. I don't trust my gaming computer precisely for this reason. I wonder if some form of containerization/virtual machine could make it possible to both allow for this kind of kernel-level anti-cheat without exposing my whole system to untrusted code.
No. Because then you could write an aim-bot that runs in the host machine…
This may be true for a class of cheats, but it isn't broadly true. That means these category of cheats can only enhance what is already visible.
In FPS games, being able to see players position through walls is a super valuable cheat _and_ requires access to the underlying game engine (to extract player location).
Maybe because doing the math to figure out if it should be visible is handled completely on the frontend to keep latency low?
This is the equivalent of a bank API giving you everyone's information, and the frontend SPA filtering it to only the logged in user.
https://technology.riotgames.com/news/demolishing-wallhacks-...
There are also videos on Youtube demonstrating it in actions
But there is plenty when you google the key words.
Anti-Cheat on the Account-Level is meaningless. You ban a cheater? Who cares. He will back with a new Account some Minutes later. Hardware Bans are useless aswell, as they can easily be spoofed. The only effective strategy for Anti-Cheat is a Universal Identity Provider, which can proof you are a real Human with no Cheat Record.
Isn't that what you are talking about here?
You can create a new Steam account, sure, but that would start you again from scratch as a total newbie (and losing access to anything you'd bought)
I've wondered if there's some sort of economy here, i.e. someone with a ton of stolen accounts monetizing them by offering to cheat all day on your competitor's server.
Most of these accounts have security questions set, so you can't change the email and truly "take ownership" of the account, but it's good enough for cheating on servers.
Some shops even have an API for generating accounts or tokens, so your hack client can have a feature where you get a new account right from your game. Get banned, reconnect in a few seconds.
Kernel level anti-cheat is meaningless.