Valve uses Overwatch to teach AI how to hack CSGO

The developer of Counter-Strike: Global Offensive once explained how Overwatch, Trust Factor, and VAC work together to catch hacks in CSGO. We will return to GDC 2018 (Game Developers Conference, game developers conference) to review this information.

In 2018, CSGO development member John McDonald was present at GDC to present the topic “Using Deep Learning to confront hacking in CSGO”. The presentation explains in depth the analysis process and how the Valve system resists hacking in CSGO. John detailed how Valve approached this issue.

Trust Factor and Overwatch pave the way for CSGO’s VACnet

After explaining the types of cheats in CSGO, John dives into CSGO’s Trust Factor. According to him, Valve has been quietly adding the Trust Factor for 6 months before announcing it to the CSGO community.

“It goes like this: you see a post on the Steam Community forum that says ‘CSGO is full of hacks’. We’ll be looking at users’ Trust scores and they’re related to 50 accounts, of which 49 were banned for hacking,” said John McDonald.

John McDonald went on to explain that Trust Factor doesn’t actually prevent players from hacking. What it does is group people whom Valve considers highly likely to hack together, thereby minimizing the impact these potential hackers have on genuine players.

The key part of the presentation when John talked about Deep Learning technology used to combat hacking in CSGO. Deep Learning is a form of computer that learns itself and is very good at recognizing behavioral patterns. Valve’s idea is to exploit Deep Learning to detect hacks in CSGO. But before developers can use Deep Learning to combat CSGO hacking, they must find a way to teach machines to recognize suspicious behavior in the same way that humans watch overwatches.

John’s idea was to use Overwatch’s conclusions as “sample data” to train VACnet, Valve’s Deep Learning anti-cheat system. If you remember, there was a time when Overwatch was suddenly full of “spin bots” right after Valve publicly introduced VACnet in CSGO. The first wave is likely due to VACnet being set free to self-assess cases, while CSGO developers focus on finding Machine Learning algorithms based on the most obvious cheats in CSGO. In addition, VACnet supports reporting to players, ie random matches selected by the software, in addition to the cases that are already infected with Overwatch.

Matchmaking issues in CSGO could be the next evolution of VACnet

While matchmaking is said to be “hacked” by some gamers, the large number of CSGO hackers may be related to the new Trusted Mode (which is something the CSGO team doesn’t want to use). But with an anti-hacking effort from Valorant, VAC was forced to roll up its sleeves to not be left behind. CSGO is forced to use a stronger Trusted Mode along with what John McDonald said, that is the future of anti-cheat CSGO.

In the future, the CSGO developer wants to show VACnet a player’s demo from start to finish, John explained, to train it to recognize hacks that are arguably harder to spot, like wall hacks. To do this and expand the VACnet database, developers can relax limits on player trust for a wider variety of reports, allowing VACnet to expand its functionality.

This way, with CSGO’s Trust Mode, the game can become cleaner. It is also the reason some players have been affected in matchmaking in recent months.

You can watch John McDonald (Valve)’s full presentation at GDC below:

You can find other related articles here: