3kliksphilip
1 HN recommender · recommended in 6 months, 2019–2023
13:04 · 574K views · Apr 10, 2018
3kliksphilip summarizes Valve's GDC talk on VACNet, a deep-learning system that flags CS:GO cheaters and feeds the Overwatch review queue. It covers Trust Factor, VAC, and how the data is collected and analyzed.
HN cites it as a source for how Valve uses machine learning on player behavior to detect cheating.
6 comments from Hacker News
If you're interested, Valve uses deep learning model called VACnet to "watch" every player on their servers[0][1]. If any dodgy gameplay is detected, it gets submitted into a moderation queue called Overwatch[2], which is basically delegates the decisions to ban cheaters community. It's really cool how they've gamified this. [0] [LINK] [1] this video [2] [LINK]
That's not what I said - my point is that companies with a relaxed security baseline (ie. not having a dedicated team of people tasked with investigating cheats and improving their anti-cheat system) tend to invite an increased number of cheaters into their games, and I back this up with real events. - These companies will send cease-and-desist letters to cheat makers instead of working to detect their cheats (Joduska.me ceased operations due to a Riot C&D, and Bossland GmbH was actually sued and lost). If these cheats were instead detected and users banned on a continuous basis, no lawsuits would be needed. - CS:GO implemented (as in, almost a decade ago) an "Overwatch" review system to allow players to review potential cheaters and have a consensus on whether or not cheating is likely. Valve is knows to be very relaxed on VAC - and knowing the company culture @ Valve being what it is, these ban waves probably only happen when one of the seniors @ Valve gets killed by a cheater in-game themselves and tasks someone with implementing some detection mechanism into VAC. Only recently (~4 years ago) has valve supposedly started to incorporate more advanced automatic detection and punishment systems[0], but at the time it wasn't doing anything about wallhacks and at auto-headshot cheats tend to be pretty good at adding enough noise to make it look very close to what professionals can do and thus these cases still end up in the overwatch queue (and note that this system probably ha
Worth noting that the VAC games on that list, CS:GO, TF2, and Dota 2, have effectively given up on actually detecting and preventing cheaters and now rely on server-based or community-consensus methods of finding and punishing cheaters (with TF2 being the exception with how little maintenance Valve performs on it). Both CS:GO and Dota 2 have a report+overwatch system where users watch a replay to determine if someone is exhibiting suspicious cheating behavior. The server side of this is VACnet where it uses heuristics like mouse movement or [in Dota] clicking out of regular camera bounds to detect these cheaters and expedite them to the Overwatch queue. You can readily download the biggest script client for Dota 2, and VAC ban waves for using it have been unheard of for years now (you only get an in-game ban when enough other players verify the cheating via Overwatch). this video [LINK]
There exist good solutions, but none of these companies have the competence or will to actually solve the problem. It’s security theatre a la malware. “Hey everyone install this rootkit to pretend like we’re preventing cheaters. Don’t worry about the keylogging and screenshot uploading.” What’s most surprising is that the vast majority of gamers aren’t outraged by this at all, despite repeated demonstrations of untrustworthiness by anticheat companies. this video
this video Valve is already using machine learning to analyze player mouse movements across all games to look for "inhuman" patterns and send those directly to the community-staffed manual review process called overwatch.