Jump to content



Permanent bans
When issuing bans through an RCON provider, such as Battlemetrics, it is rather easy to bypass bans. All a player needs is a new Steam account and a new IP address.
RCON tools have access to a limited amount of information, all they can access is your IP, Steam UID and Steam account information (account age, number of games, etc).

With scripted solutions, you also have access to profileNamespace (which stores data on your PC, regardless of Steam account) and hardware devices (GFX device, input devices, etc). Using all this data, each player has a profile created for them. The profile includes information such as which game settings, keybindings, ISPs/VPNs, and devices are being used.

As new players connect, we can evaluate a risk factor for them based on their Steam account and how many times they’ve played before. If the risk factor is high enough we can check if there are any matching profiles. We feed a list of all player profiles into GCP AutoML Tables, which will decide if there is a high enough statistical probability of an alt account. Recently banned accounts are given a boosted mean score, meaning new accounts are more likely to get linked to already banned accounts.

When training an AI, the biggest issue is the dataset. For satisfactory results, you need to maximize relevant data and minimize irrelevant/incorrect data.
Our solution to maximizing data relevancy is scrubbing out all duplicate data, so each identifier has only one instance.
After our data has been cleaned and formatted properly, it is fed into GCP AutoML Tables. For each new connection, we call the AutoML Tables API, which will return the probability of an account match. If the match probability reaches a certain threshold, the matched accounts are linked.
With AI-based solutions, we can easily identify players based on their VPN/internet providers, IP locations and game configuration options. Our system is similar to existing fingerprinting technologies but more flexible.

We believe the AI is able to detect 80-95% of alt accounts. Currently, about 50% of all global bans are automatic (compared to less than 19% on infiSTAR). Getting accurate data on how many alt accounts we have missed is a bit tricky since cheaters don’t really want to hand out their new accounts’ info.

The AI automatic alt account ban system is the most advanced in Arma and we are continuing to improve it.
Our future plans include providing a risk factor profile in the antihack dashboard, so server admins can set their own risk thresholds. We are also automatically pushing new and updated datasets to the learning algorithm every week, so the prediction model will continue to improve.

Alakategorioita ei löydetty