.svg)
CapsuleAI's Orca automates accurate Gameplay Review to better detect cheating and future-proof against Machine Learning (ML) cheats. This frees up game devs to spend their time where they really want - making great games!
Let's see how it works 🧐
Orca cheat detection analyses player behaviour to determine whether cheating has occurred. Detecting cheats in this way completely avoids considering the cheat mechanism, because the result of using a cheat will be observable, regardless of how it is implemented. Orca addresses all mechanisms for cheating because it doesn't care about the mechanism.
Traditional anti-cheat, on the other hand, works by tackling specific mechanisms which enable cheating. Obstruction is the main method, which attempts to prevent any tampering with a game as it runs. Obstruction relies on tampering being detectable, making it vulnerable to innovative mechanisms for hiding cheats. This means obstruction requires constant maintenance to stay relevant. It also often causes frustration for players of games, with intrusive or clunky implementations and false-positive detections.
A recent development in cheating exemplifies how the behavioural approach is better than traditional obstruction. External cheat devices change player inputs without ever interacting with a game as it runs. Obstruction simply cannot prevent this. Orca would detect cheating utilising this, or any other method.
An automated process can scale review to any number of games at a much lower cost than if done manually, enabling consistently accurate cheat detection across an entire playerbase. Machine Learning makes it possible to achieve better than human levels of accuracy in cheat detection, by taking advantage of large numbers of examples distinguishing patterns of normal player behaviour from cheating behaviour. Orca is even capable of bootstrapping its own learning when labeled data is not available.
Many games make use of a manual review pipeline to detect cheaters, CSGO uses Valve's Overwatch for example. These systems work by attempting to find moments of gameplay where cheating may have occurred, with a human reviewer making a judgement. The fundamental issue with human review is that it is very expensive, which prevents it from scaling to all gameplay. There will never be enough reviewers for every moment of gameplay to be individually reviewed, so filtering out only relevant moments becomes very important.
Cheating is only becoming more and more prevalent, increasing the pressure on review systems. Only automated review can scale to the volume of online games being played.
Advances in ML have also meant advances in cheat implementations. As seen earlier, ML cheats enable off device cheat implementations, which entirely avoid obstruction anti-cheat. Even when cheats remain on device, techniques such as computer vision interact indirectly with the game. As cheat mechanisms become subtler and harder to detect, so the invasive impact of obstruction techniques like device scanning become worse. These methods aren't just theoretical, they are available to players right now.
But no matter the method of cheating, new or old, they are all exposed to behavioural analysis. No cheat can avoid behavioural review because they cannot avoid appearing unusual, playing the game unlike any normal player. Automated review guarantees that all cheaters will be seen.
The good news is that the future of ML based automated cheat detection is already here. Orca automated behavioural review provides a scalable, future-proof and accurate means to detect cheaters in games.
CapsuleAI's Orca is an automated review system utilising machine learning to detect patterns of behaviour that show cheats have been used. Orca takes game event (AKA message) data in a unified format as input, and returns time bounds for when a player cheating has been detected. Because Orca works with game event data directly, teams and companies can easily stand up and automate their game review pipeline.
In testing using labelled and unlabelled real game event data from Assault Cube matches, Orca was able to achieve accuracies of up to 97.4%. These results are accurate enough to be actioned automatically, and can also be used to focus existing human review capability on problem players, increasing efficiency and reducing cheating simultaneously.
We are looking for partners for our Alpha programme to further validate the system in realistic contexts. If you would like to get involved you can get in touch here!
.svg)
