Real Aimbot

[Logan]

For any admins, I have been compiling a list of IPs that I blackhole of known aimbot users. Most recent one(last night):

98.164.90.240

I keep a list of IPs of aimbot users or bitchasses. If anyone running a server wants a list, I can make it available.

Nice work Mar. I havn't seen any aimbot users at DIGICHALK for some time now I still havn't been able to try out the banning process.

[Marquis De Sade]

New aimbot user:

IP: 96.54.239.171 (Nanaimo, near/in Vancouver, Canada)

Confirmed via spectating, and also by the words of the aimbot dirtbag. The nick was 'F-ing C*nt' but completely spelled out - seems there is some censoring of certain words on forum

say: UnnamedPlayer: bot
say: F-ing C*nt: obviously
say: UnnamedPlayer: bye bye
say: ^1Nudist^3Noob: NUDISM FTW!!!
say: MarquisDeSade: blackholing
say: MarquisDeSade: 96.54.239.171 is ip
say: F-ing C*nt: 4 now
say: MarquisDeSade: ill block the whole network of your isp
say: MarquisDeSade: bye bye dipsh*t
say: ^1Punk: :p
say: MarquisDeSade: shawcable user fyi

[Logan]

Nice work MAR.
I guess ciggaweed gets all the aimbot users these days because I still havn't seen one at digichalk in a while.
haha, that's fine with me. :p

[Marquis De Sade]

New one:

IP: 98.212.3.179

Name: . TEE HEE

[GrosBedo]

Just a thinking about a general anti-cheat system :

Instead of trying to outsmart technically the aimbots by spotting their traces and code injections, why not just do a behavioural test ? I mean, we all know how to spot an aimbot, even if some are more obvious than others, and this could be reproduced by a software, for example with a statistical and heuristic approach, we could spot with a fuzzy result a certain percentage of certitude about a player cheating or not. Plus, this approach would not be avoidable by the clients since it would be done server-side.

For example, some acts would have more weight than others, like correcting each MG bullets.

All it would require to integrate a system like this is a little more CPU processing serverside. Nothing clientside.

The goal is not to automatically kick or ban those users, but at least report them in a log, and if possible, attach a demo of some duration for later review by an admin who can then do the right procedure (drop it, warning, kick, temp ban, perm ban, refine anticheat settings).

On the other side, I know it's possible to make very smart and human-life aimbots, but isn't our goal to avoid non-human skills ? As long as an aimbot is, at worst, like humans, then the problem is much lower (and anyway aim is only one of many other skills a good player should have, alone it can't surpass a good human player).

What do you think about it guys ?

Note : I can make an algorithmic or Python generic code as a basis, but Im not experienced enough with C++ and the ioquake3 sourcecode to cleanly do it.

[HelloKitty!]

I'm pretty sure that any such algorithm would not be reliable at all.

Spotting obvious aimbots is not terribly useful. You can spot them even without specing. Spotting non-obvious aimbots is tricky even for people.

[GrosBedo]

I'm pretty sure that any such algorithm would not be reliable at all.

Spotting obvious aimbots is not terribly useful. You can spot them even without specing. Spotting non-obvious aimbots is tricky even for people.

Details is the devil. When you sum up several subtle weird behaviour, you would theoretically end up with a quite good detection, even with smarter aimbots. Don't forget that a computer can do more precise statistics that a human.

As I said, the goal would be primarily to spot and in the end prevent non-human skilled aimbots, but this doesn't mean that only the obvious aimbots would be spotted !

There's no magic solution :  you either lower down the risk, but accept that you can't fully eradicate it, or you do a constant race to keep up with the risk. As you want, personnally I don't want to spend my life for a useless cause. I know hackers XD

[Falkland]

I'm pretty sure that any such algorithm would not be reliable at all.
[...] Spotting non-obvious aimbots is tricky even for people.

It could be reliable indeed : it depends only of the accuracy of the analysis because there's always a pseudo-randomness in such those bots .

You cannot code any kind of aimbot that has contemporaneously an acceptable randomness ( for movements or tactics or hits ) and a response time of milliseconds and that is not running on a supercomputer.

If you ever had some kind of experience with cryptography , you should know that to avoid pseudo-randomness effects ( in other words , prediction ) you need some kind of entropy , an initial seed and cycling inputs many times to obtain something enough close to randomness, but it requires some seconds of time on a nowadays CPU, or even more.

So even the most sophisticated aimbot operates always in the same way because it naturally lacks of randomness.

About weapons and specifically the MG ... I've noticed 2 kinds of bots tyring to "un-spread" bullets :
- bots that uses a "burst" fire algorithm ( in which the burst time is lower than the rapid fire time ) to avoid the natural game bullet's spread algorithm . Even if the autoaim is configured to drop 2 bullets every 3 fired ones ( for example ) , the accuracy is always around 33%.

- bots that automaticily correct the spread with special prediction code client-side that correct aim before the next bullet is fired. This kind of bots are more easy to spot because usually the player under attack feels like he can't move or have some difficults to move. Usually those kind of bots needs a constant game world consistency and doesn't like packets loss , so they usually crash in few minutes in case of systemic, even artificially produced ,  packet loss

About the aim , the aimbot always aim at player with the selected ( usually by cvars ) options : aiming through walls or not , aiming at the closest enemy or over the enemy under crosshair or over the enemy close to the ch , aiming with a HIGH FOV ( easy to spot ) or with a LOW FOV , aiming at the player with the lowest HP points ,  etc ... for example , if you have two enemies close to each others and fightimg themselves , it's natural to aim in the middle with RL to try hit both while a bot usually aims possibly with an insta weapon over the first one (the closest or the furthest or the player with the lowest health ... ) and _THEN_ over the second one .

Another aspect to be considered is logging of the client input ( key pressed , MOUSE clicks and movements ). Usually bots'  automations don't make use of the input subsystem ( eg , the autoaim system which points the ch over the enemy [ +left or +right ] but without logging a mouse movement or a key pressed ; at least it could log a LEFT MOUSE CLICK [+attack] )

[GrosBedo]

It could be reliable indeed : it depends only of the accuracy of the analysis because there's always a pseudo-randomness in such those bots .

You cannot code any kind of aimbot that has contemporaneously an acceptable randomness ( for movements or tactics or hits ) and a response time of milliseconds and that is not running on a supercomputer.

If you ever had some kind of experience with cryptography , you should know that to avoid pseudo-randomness effects ( in other words , prediction ) you need some kind of entropy , an initial seed and cycling inputs many times to obtain something enough close to randomness, but it requires some seconds of time on a nowadays CPU, or even more.

So even the most sophisticated aimbot operates always in the same way because it naturally lacks of randomness.

Um Im not sure what time randomness has anything to do here, but indeed bots can't be too much random, else it would lower drastically their skills (isn't this what we want ?).

About weapons and specifically the MG ... I've noticed 2 kinds of bots tyring to "un-spread" bullets :
- bots that uses a "burst" fire algorithm ( in which the burst time is lower than the rapid fire time ) to avoid the natural game bullet's spread algorithm . Even if the autoaim is configured to drop 2 bullets every 3 fired ones ( for example ) , the accuracy is always around 33%.

- bots that automaticily correct the spread with special prediction code client-side that correct aim before the next bullet is fired. This kind of bots are more easy to spot because usually the player under attack feels like he can't move or have some difficults to move. Usually those kind of bots needs a constant game world consistency and doesn't like packets loss , so they usually crash in few minutes in case of systemic, even artificially produced ,  packet loss

About the aim , the aimbot always aim at player with the selected ( usually by cvars ) options : aiming through walls or not , aiming at the closest enemy or over the enemy under crosshair or over the enemy close to the ch , aiming with a HIGH FOV ( easy to spot ) or with a LOW FOV , aiming at the player with the lowest HP points ,  etc ... for example , if you have two enemies close to each others and fightimg themselves , it's natural to aim in the middle with RL to try hit both while a bot usually aims possibly with an insta weapon over the first one (the closest or the furthest or the player with the lowest health ... ) and _THEN_ over the second one .

Thank's for these interesting inputs, this is the type of detail that an automated anti-cheat system can look for, and averaging the results on a certain time duration would permit to avoid false positive, plus a mediuming (don't remember how it is mathematically called, I mean the calculation of the highest values) would still permit to spot smart hacks that drop skills for some period of time to average their results.

Anyway, this journey would require more than empiric observations, so we would have to test the currently available hacks to get a finer picture of what is running currently.

But if done well, this system should be adjustable for future needs with only some parameters tweaking.

Another aspect to be considered is logging of the client input ( key pressed , MOUSE clicks and movements ). Usually bots'  automations don't make use of the input subsystem ( eg , the autoaim system which points the ch over the enemy [ +left or +right ] but without logging a mouse movement or a key pressed ; at least it could log a LEFT MOUSE CLICK [+attack] )

I don't think this is a solution, because this can be easily simulated. This would only eliminates current low-level hacks, but in the future they will all bypass this protection. Plus this would require client-side modification in the engine, I don't think this is good enough compared to the efforts it would need.

[GrosBedo]

Another thing : adding client-side tests would require more bandwidth usage since more infos would have to be transferred to the server (plus the risk that they get faked by hackers), while on the other side a server-side behavioural anti-cheat system would require no more bandwidth usage, everything is already available, so that's why it would require only some more CPU usage on the server, it would be completely transparent to users.

[RMF]

I would rather care about MITM attacks, altered client executables or any other way of manipulating the transmitted data than caring about sending a few extra bits to indicate if it detected a cheat yes||no.

[HelloKitty!]

the problem is that EVERY single symptom of an aimbot that I've ever heard of is very common with real players. Here are some of the ones which were  cited on here:

- shooting without mouse movement
- flick rails
- railing in mid-air and/or while strafejumping
- /kill
- smileys when falling into void
- suicides due to hitting a wall with a rocket

The problem is, I do ALL of those, and I certainly don't use an aimbot. I wouldn't even know where to look for one. If I used an aimbot, my aim would not suck as much.

Even any combination of those is not a reliable indicator that someone is using an aimbot.

Any statistical analysis will not be conclusive. You can model the behaviour of KNOWN BOTs and detect that, but you cannot model the behaviour of all human  players and then detect anything that deviates from it and label it as a bot. Detecting obvious bots is easy. Detecting subtle ones is not, and I'm afraid that clever detection algorithms will not help there. It's not that the algorithms are bad, it's that the problem is too difficult.

As machine learning people know, garbage in = garbage out. Computers are not magic, they are simply good at number-crunching. And when it comes to analysing human behaviour, they are usually far inferior to human beings. If you don't have representative features that can discriminate between your classes (cheater vs. non-cheater), then you cannot possibly learn a classifier to distinguish between them. Even a complex combination of many factors will only give you correlations which will not approach 100% accuracy. You'll inevitably flag a bunch of honest players as cheaters, and really, you don't need software to do this -- people are crying "bot" at decent players just fine without any clever algorithms.

If anyone makes a close to 100% accurate bot-detector program based on behavioural analysis, I'll eat my hat. It's a pipe dream.

[GrosBedo]

I would rather care about MITM attacks, altered client executables or any other way of manipulating the transmitted data than caring about sending a few extra bits to indicate if it detected a cheat yes||no.

That's why I say that it's useless to do a client-side test. Anyway, about the bandwidth, it would require more to send each keystroke and each mouse click. This would be useless.

the problem is that EVERY single symptom of an aimbot that I've ever heard of is very common with real players. Here are some of the ones which were  cited on here...

Because you've heard noobs complaining. Im not a noob. And the list you've given is pretty much vague, this is totally normal that if you use it as a basis to spot aimbots, you can go nowhere.

I know aimbots because I was in the scene some years ago, I didn't really make one but I liked to analyze how they were done, particularly because they were a plague in Enemy Territory and the only way to get back a kind of fairness was to use one yourself.

Now that I have more knowledge in IA and statistical analysis, I would like to propose a new way to protect against this threat rather than fight fire with fire.

Any statistical analysis will not be conclusive.

Sorry to be rough, but this sentence just shows how ignorant you are about mathematics and computer engineering. No, this is not magic : this is science. And this is possible. So many things are possible with science, like flying, becoming invisible (now almost true), making replicable machines, 3D manipulable holograms and seeing through clothes (yum).

Spotting a bot by behaviour is just not beyond possibility. And if your imaginary stops at this point, go in museum and conferences more to wider your vision.

Computers are not magic, they are simply good at number-crunching. And when it comes to analysing human behaviour, they are usually far inferior to human beings.

Is that why a computer already knows what is the next hit song that you will love ?

http://en.wikipedia.org/wiki/Hit_Song_Science

Sorry to deceive you, but you, I, and any other on this forum, are far more predictable than you believe. We simply are machines.

You can model the behaviour of KNOWN BOTs and detect that, but you cannot model the behaviour of all human  players and then detect anything that deviates from it and label it as a bot. Detecting obvious bots is easy. Detecting subtle ones is not, and I'm afraid that clever detection algorithms will not help there. It's not that the algorithms are bad, it's that the problem is too difficult.

Exactly, I will detect everything that is NOT a human behaviour. This has to be done in a systemic (general) way and a detailed one, both are necessary. For example, if you have something near 100% accuracy at the end of a match, and you killed a fair amount, then this raises the probability that you cheat. If you correct each bullets, or exactly each bullets of 3, this will be recorded, and accounts for you to cheat.

Any one of those behaviours alone will not tag you as cheater, but they will raise a certain percentage of cheating, thus achieving a fuzzy logic.

And, as I said, the goal is not to automatically ban cheaters, but rather spot them for future reviewing, so the best would be to autorecord a demo of players reaching a certain level of cheating probability percentage.

---------------------------------------------------

Ok so, who's with me ?

I need to collaborate with someone that knows well C++ programming, and has at least a basic knowledge of how the game works and knows about the sources. I don't need someone that really knows the engine by heart, I can give sources, explain roughly how they work (that's enough, search is your friend after), and give links for more infos describing the different parts of the engine.

We don't even need to modify the binaries, I think that this can be achieved in a mod. Later it could be integrated, so it could be used with other mods, but for the moment a mod would suffice.

A modder interested into revolutionizing anti-cheats system for opensource games here ?

[andrewj]

Spotting a bot by behaviour is just not beyond possibility.

It is definitely beyond possibility.

Any analysis you do will have a certain threshold where the best player doesn't cross it, but your run-of-the-mill aimbot would (such as accuracy with MG).  This threshold needs to be pretty high, as some people are very very good.  Hence all a bot maker needs to do is make sure his algorithm never crosses that threshold.

Your system would work to eliminate the super fantastic (obvious) bots, but they will then be replaced by slightly more sophisticated bots which are indistinguishable from the very best players.

It's been said many times before, and I think it's true: cheating prevention with FOSS multiplayer games is a problem whose solution is not a technical one but a social/community one, where the individuals who cheat are detected by peer review and banned from the servers.

[GrosBedo]

Any analysis you do will have a certain threshold where the best player doesn't cross it, but your run-of-the-mill aimbot would (such as accuracy with MG).  This threshold needs to be pretty high, as some people are very very good.  Hence all a bot maker needs to do is make sure his algorithm never crosses that threshold.

First, the thresold will be configurable, secondly they will not really be threshold, and it will not be as obvious as "accuracy with MG". It will look at subtle details and patterns a human CANNOT do, and will distinguish between chance and computation with mathematical tools.

Your system would work to eliminate the super fantastic (obvious) bots, but they will then be replaced by slightly more sophisticated bots which are indistinguishable from the very best players.

Even sophisticated bots CAN be distingued from the very best players. I once saw one like this : in normal speed, you couldn't spot it, but if you watched the demo in slow motion, you could easily tell that it was a bot. The same happened otherwise (from normal speed the player was like an aimbotter, but at slow pace, it was clear he wasn't).

This approach would for sure reduce greatly the use of aimbots (at least it will spot them, admins do whatever they want after with the datas), this is the goal ! I cannot assure that ALL cheating will be banned for now and forever. We could do that too, but it would require an extended tracking of the evolution of the current hacking scene : I don't want to loose my time with that, I'm not paid for that.

But, if we could get a tool that is intelligent, adaptable, easily extendable (opensource of course), and tweakable, I think that it would for sure help quite a lot.

The only flaw of this system that I can see, is an aimbot that is EXACTLY like a human, I mean even in the very subtle and unconscious features of the human motions (and it's hard to reproduce, look at the BrainWorks blog). But this not as bad as it looks :

1- it's almost impossible to do that with an aimbot, because of technical restraints : BrainWorks is a mod, it has access to ALL informations and features of the engine, and aimbot just don't (first it's client-side, and second it's injecting into the code, thus it can't really add or modify too much the features to be able to reproduce a human).

2- even if someone do it (sky isn't the limit anymore), then the problem is still reduced because it CANNOT be better as a human can be. Where is the problem then ? And anyway, as I said, aim doesn't do the whole thing. Current "perfect" aimbots can be beaten by pro players (human brain still calculates more things than a bot), so if we even reduce the skills of the bots to a "human" level, they won't really be a threat anymore.

It's been said many times before, and I think it's true: cheating prevention with FOSS multiplayer games is a problem whose solution is not a technical one but a social/community one, where the individuals who cheat are detected by peer review and banned from the servers.

I agree, but as I said and I will resay again :

THIS IS NOT THE ULTIMATE ANTI-CHEAT SYSTEM ! THIS WILL JUST REDUCE THE USE OF AIMBOTS !

Ok now ? There is no full bulletproof solutions, anywhere in the life, same goes for computers. But if we can reduce the risk, then it's good, and we can do quite a lot with this system.

Now as you want, personnally I don't have a problem with aimbotters : I can play against them, and there aren't really many, this is a myth. And if an aimbotter is spotted, you can still callvote kick him XD

Personnally I don't care, but instead of just basing your assumptions over just some rough idea of what an aimbot is, I propose to do use a full statistical approach using real mathematics, and not just some obscure reasoning like "he never miss a bullet - he knows everywhere I am - he did some flick shots"...

[Cacatoes]

No, this is not magic : this is science. And this is possible. So many things are possible with science, like flying, becoming invisible (now almost true), making replicable machines, 3D manipulable holograms and seeing through clothes (yum).

Hmm, great stuff, but is science able to protect some DVD from being copied, or some software to be cracked ? Seems a rather simple task compared to distinguishing human/bot moves.

Sorry to deceive you, but you, I, and any other on this forum, are far more predictable than you believe. We simply are machines.

Oh duck ... you seem so confident with statistics. You talk about predicting musical tastes of people who are already lobotomized and accustomed to hear only one kind of music. It's not like these songs had serious artistic complexity, and it's not like people were liking it because of how it sounds rather than because of marketing. They talk about HIT SONGS, it's not a coincidence, they don't talk about any masterpieces.

I mostly agree with andrew.

If you implement rules for anti-aimbot, then these rules would be reused inside the aimbot so that it takes them into account and bypass them.

[GrosBedo]

Hmm, great stuff, but is science able to protect some DVD from being copied, or some software to be cracked ? Seems a rather simple task compared to distinguishing human/bot moves.

Ok, let me repeat once more : this system is NOT perfect ! Not any system can be ! There are always drawbacks. And your presumption is false : try to crack Trackmania 2 by your own, just for fun. Good luck.

I propose to reduce greatly the risk of aimbots usage, not to suppress it altogether. But it will be so hard to bypass it that it will be unlikely that someone goes through.

Improbable thoughts ? Take a look at cryptanalysis, yes the thing that let you crypt your confidential documents and even the ones of secret services : they ARE bypassable, but they would take so much calculation and efforts that it would take years to crack a good password with a good crytpalgorithm, thus protecting effectively (as far as possible in fact) the crypted documents.

There are no bulletproof system, go over it. But we however CAN do something about it.

Edit: Ah and secondly, here the goal is not to make a stupid protection for a content that should be available to you (DRM), but to analyze some specific behaviours that are different from humans.

No seriously, now you have to decide guys : are we machine, or not ? Because, if we aren't machines, then there are features that distinguishes us ! Because if bots are undistinguishable, that means these machines are exactly like us !

Machine or not ? Paradox ?

You talk about predicting musical tastes of people who are already lobotomized and accustomed to hear only one kind of music. It's not like these songs had serious artistic complexity, and it's not like people were liking it because of how it sounds rather than because of marketing. They talk about HIT SONGS, it's not a coincidence, they don't talk about any masterpieces.

Mozart and other classical songs are integrated into the algorithm (at least the database that the algo processed).

No really, I know it's hard to believe, I can imagine your nice childhood you've had while dreaming of Mickey and his friends, and watching Goldorack on TV (don't take me wrong, I love Goldorack and Mickey, but I don't believe we're living in their worlds).

I won't do you a full course about how predictable and manipulable you are, we'll simply do a simple test : go out naked for one full hour shouting "I LOVE BRITNEY SPEARS !!!!". Ok, too hard ? You fear that police arrest you ? Then forget the first bit, just go out and shout "I LOVE BRITNEY SPEARS !!!!" accross the town for an hour. You don't have any chance to get arrested just for that.

Here's the conclusion for you : you can't. Why ? Because your choices (free will) is just a ponderated selection over your possible choices ponderated by your past experience. This is just beyond your possible modelisation of the real. That's all. Someone other could do it.

Now, I don't need you to believe me, after all denying is the most predictable human reaction.

If you implement rules for anti-aimbot, then these rules would be reused inside the aimbot so that it takes them into account and bypass them.

If they have server-side access, I agree. Else, no.

Opensourcing a system doesn't mean that it's for sure breakable. Look Linux. And Microsoft Windows is closed source.

---------------------------

Anyway I propose to make it and prove/disprove that it works by trying. I don't just argue some features Id like the dev to put in !

If you are all afraid to try, then go play and have fun.

PS : No, Im not against your replies guys, I thank you for your interesting inputs, but the "It cannot work because it's not possible/bots are too smart/everyone tried and failed/this is too complex for me" arguments are NOT valid. Please first try to imagine this system from what I say, and if there are things you find obscure or case that don't fit, please tell me so and Ill try to answer. But please, imagine it is possible, Im tired to argue if it is or not, because it won't be settled until it's done anyway.

PS2 : Please try to really understand that there is not point to try to make an unbreakable system, this simply can't exist (try to give me one example). When you'll have understand that, it's more easy to think of a system that is not unviolable, but efficient enough for its purpose.

[RMF]

and seeing through clothes

oh hell yeah im with you man!

edit: ok read through the rest of the topic. What I think is that we just shouldn't look for a solution to eliminate aimbotters. We should just make it hard enough. When we detect for example the obvious things like when the crosshair is always less than x pixels off a player, when the accuracy is either very low or very high with certain weapons (the king orgy client hook for example can't hit a thing with rockets unless its lucky), etc. Sure like caca said "these rules would be reused inside the aimbot so that it takes them into account and bypass them". But it will kill frag most of the aimbots around. Also often code from other games which are alike can be reused for oa, when we make cheatprotection unlike familiar games they have to rewrite the code and it is less likely they port a bot.
Another thing we should target is the exploit allowing cheaters to join servers where they are banned. I never saw it or heard of it here, but almost every time I talk with an aimbotter (or sometimes some die-hard campers too) they say that they will join anyway ban or not due that exploit there is.

[GrosBedo]

and seeing through clothes

oh hell yeah im with you man!

You see my point ? That's science !

[HelloKitty!]

Sorry to be rough, but this sentence just shows how ignorant you are about mathematics and computer engineering.

I'm sorry to be blunt, but I actually do that for a living.

When you make a foolproof bot detection system, let me know, and I'll buy you a beer.

[RMF]

I got one. It also wrote what you are reading now.. ye i mean myself
So, the beer?

[HelloKitty!]

Yeah, but you didn't MAKE that system. Your parents did.

[GrosBedo]

Sorry to be rough, but this sentence just shows how ignorant you are about mathematics and computer engineering.

I'm sorry to be blunt, but I actually do that for a living.

Do you mean computer engineering or mathematics ?

When you make a foolproof bot detection system, let me know, and I'll buy you a beer.

I don't like beer. Vodka ?

[HelloKitty!]

Do you mean computer engineering or mathematics ?

Machine learning.

Behaviour recognition, to be more specific

[RMF]

Yeah, but you didn't MAKE that system. Your parents did.

Give them a beer! Btw it's my fathers birthday today (no joke)

Ok but let's get ontopic again.
So we're first going to try to get only really obvious bots detected? I guess we can't do a lot more and it'd be the first step anyway.
I am newb at scripting/programming other things than basic programs not that interactive with others. No experience with C or whatever ioq3/oa is built in so I guess I can't help that much. Some other experts willing to help?