Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive

From what I know, VAC (Valve Anti Cheat) just looks for processes running on the system and detects injections into CSs memory. Then, for CS:GO specifically there's Overwatch, in which players look at other reported players' gameplay to determine whether they were cheating, and VacNET which is a machine learning system trained on the data from Overwatch to detect aimbots that way. There's a really good talk that someone from valve gave about 3 years ago[0]. The bigger problem is that even with input recognition, one of the biggest problems are wallhacks, meaning you can see other players through walls which is an advantage that's almost as large as aimbotting in tactical shooters like CS. [0]: [VIDEO]

John McDonald did a talk at GDC 2018 about how Valve was starting to integrate deep learning into their server-side cheat detection. IIRC, some of the process was automated, some was player-reported, and some was done through human review. As a gamer averse to both cheating and spyware, I found this very appealing. I wonder how their systems have evolved since then. The original youtube clip (v=ObhK8lUfIlc) has been made private, but I think this is a copy of the same video: [VIDEO]

Valve do this, here's a good talk: Robocalypse Now: Using Deep Learning to Combat Cheating in Counter-Strike: Global Offensive [VIDEO]

As far as I'm aware Valve does use ML (called VACnet) to detect cheaters. At the time of the GDC talk[0] it was only implemented for CSGO but it could've expanded since then [0] [VIDEO]

Or server-side automated review using machine learning models, which is what Valve is trying with "vacnet" (with then manual review to confirm the model's results are accurate - something most tech companies should learn from) Great talk on it here: [VIDEO]

Valve has been at this for a long time... They started with a system that lets players review games and try to understand if a player is a cheat or not. Games are anonymized, and you get to rewatch the replay to try to make a determination They can then use that data to train their NN to help detect which players / games should be fed into this manual review process I assume now the product is probably good enough that it's ranking reviewers too to determine how good reviewers are at detecting if there's cheats or not. It's pure genius Here's a GDC talk you may find interesting: > In this 2018 GDC session, Valve's John McDonald discusses how Valve has utilized Deep Learning to combat cheating in Counter-Strike: Global Offensive. [VIDEO]

Some players will also be convinced that you're cheating if you play the game really really well, and they will get upset. CS:GO (now CS2) has a very fascinating way of determining if someone is cheating. A ML based heuristic that is constantly being retrained, that can accurately judge whether someone is actually cheating or not based off their Overwatch (not the game) replay system. [VIDEO]

What an interesting day when you see a site you've worked on for the past 2 (3?) years get posted to HN! Except I tried submitting this site years ago when I had just finished it, but it did not seem like HN was that interested at the time, and I don't blame them. It was very niche and video game related, and the site also looked a lot worse. It's come a long way to the point where there where I collaborated with someone else to do a redesign, which I think has done great for the project at large. I originally created the site as a way to track which games would be supported on Linux, since at the time the Steam Deck was releasing, and some games were turning to support it. And it has since blossomed into a larger project, which some other tools even pull from! I would have never even imagined that when I first started making this. I do want to address something I see being talked about in the comments, which is the fact people say that anti-cheats are snake oil, or useless. This is a big misunderstanding, and I feel like those more technically inclined should understand that anti-cheat is a "defense-in-depth" type of approach. Where it is just one of many lines of defense. Some anti-cheats are pretty useless, and don't do much, but some actually do try and protect the game you're playing. But, just like DRM, it can be cracked, and that's why it's more of a constant arms race, rather than a one and done thing. I'm writing out a longer post about this for the future, but just

Unfortunately the Overwatch system has been disabled for years, and we currently have no reason to think it will be reopened for CS2. I take this opportunity to share this great talk about Valve's usage of deep learning to fight cheating in CS:GO: [VIDEO]

Woah, cool to see a bit of how it all works. Seems similar to basic antivirus signature checking? As a side note, a few years back, they had a GDC talk about a (then-new) machine learning based approach to catching cheaters. Regardless of how it worked out, I found it fascinating. The presenter delivered the problem and solution approach so well I'd still like to work on something like this some time. I think they're still updating this for CS2, so it would be really cool to see how far it's gotten now. [VIDEO]

this is already running in production for Counter-Strike since 2018 [0][1]. to be honest, it isn’t particularly good - all serious CS2 games operate on a third party provider with a kernel-level anti-cheat. also, the cs2 update banned people for spinning their mouse too fast [2]. [0]: [LINK] [1]: [VIDEO] [2]: [LINK]