Denuvo DRM reportedly fully cracked open, 2K apparently fights back with online checks

I could not care less for Denuvo. Publisher that treat me without respect will not earn my money. That includes games shipped with Denuvo as well as with KLAC.

Quoting: sarmadMultiplayer games should depend on server-side AI based monitoring

Agree with "server-side", disagree with "AI based monitoring". I would probably one of the first people getting banned for not hacking, because I play games like games and not guided (as 98% of all players these days do). AI is discriminating based on their trainings data. It is common sense that it should not be used for surveillance and Monitoring to filter cheaters is nothing else.

Classic server-side anti-cheats are totally fine and good enough to create a nice gaming experience for everyone. It is just more expensive to create game specific anti-cheat solutions compared to allrounder tools (where legit players pay the price). The few people that will stay unrecognized by classic server-side anti-cheat will also exist with AI anti-cheat or with KLAC or any other anti-cheat method. How do you want to detect dobbing? The goal is not to reduce cheaters to zero, but to create a gaming experience for every legit player that doesn't feel like there is any cheater and there is no much difference in capability between all these methods. However, AI-anti-cheat also banns legit players and therefor it is even as bad as KLAC.

Anyway, anti-cheat is a different topic, DRM is to prevent piracy, nothing else.

Last edited by PlayingOnLinuxphone on 29 Apr 2026 at 5:53 pm UTC

Quoting: mindedieIf perfect DRM existed,

There is. It's called a "offering a good product that's worth its money". People typically don't pirate if they don't think the pirates have the better product.

Quoting: PaldinoXThat's a shame 2K is using Midnight Suns as a testbed for this crap, nobody bought it despite being a genuinely great game and now even less people will bother with it because of this.

I got Midnight Suns for free and thought it wasn't good enough to finish (put in 23 hours). There are only a handful of mission types (kill everything, destroy thing, protect thing, etc) that you do over and over. Then the whole "social" aspect was boring to me. It felt like a cozy mini-game, that you are forced to do to get upgrades, in a tactical game.

I think the bones of the combat was good though as I like card and spatial turn-based combat. I think they could have had a much better game if they stripped the social movie bullcrap and focused more on the action. While I'm not anti-cozy, I think this game was like oil/water pulling in two different directions.

Last edited by eggrole on 29 Apr 2026 at 6:04 pm UTC

Unfortunately there's a growing probability of Denuvo moving to kernel level DRM. And if that happens, it's game over for Linux users. Not a problem specifically for me, I don't buy games with Denuvo DRM, but for the platform itself, it might be a big hit.

Quoting: doragasuUnfortunately there's a growing probability of Denuvo moving to kernel level DRM. And if that happens, it's game over for Linux users. Not a problem specifically for me, I don't buy games with Denuvo DRM, but for the platform itself, it might be a big hit.

Do you have a source? Hope this doesn't happen

And what exactly happens to legitimate players in 5-10 years time where the online check servers are killed or compromised and you can't renew your 14 day license to play the game you bought? Nothing, that's what. You just can't play it anymore. But the corporations won't care about that.

I, too, prefer attempts towards integrity, accountability and sources. If we want to be fair, we have to tolerate differing opinions. If I wanted to be in an echo chamber I would turn to an AI chatbot instead.

Quoting: PlayingOnLinuxphone

Quoting: sarmadMultiplayer games should depend on server-side AI based monitoring

Agree with "server-side", disagree with "AI based monitoring". I would probably one of the first people getting banned for not hacking, because I play games like games and not guided (as 98% of all players these days do). AI is discriminating based on their trainings data. It is common sense that it should not be used for surveillance and Monitoring to filter cheaters is nothing else.

That's a good point. I'm not sure how the AI is trained to detect cheating. But it certainly shouldn't be trained on expecting you to play like everyone else; it should be trained on more raw data that includes the whole game world, not just how the player behaves. For example, it should be able to detect that you are aiming precisely at a target behind a wall, or other similar things that are impossible without cheating.

Quoting: Caldathras

Quoting: sarmadSingle player games should have no anti-cheat protection.

As I understand it, Denuvo is not anti-cheat. Just looking at its full name, it is anti-tamper DRM -- it performs a completely different function.

Oh, my bad. I thought it was for anti-cheat.

Quoting: sarmadFor example, it should be able to detect that you are aiming precisely at a target behind a wall, or other similar things that are impossible without cheating.

That is a good example why classic algorithms are better than LLMs which are basically algorithmic lossy archive files. It is something like MP3 for all kind of data where the timestamp is the input (prompt, randomization number and so on). The important part is "lossy", because it loses information on the training process. This causes fail predictions as not being able to do correct maths. I just asked CGPT on duck.ai "What is the math result of 13/73²?" and the result was "13 / 5329 ≈ 0.002438 (rounded to 6 significant figures).", while the real numbers are 0.00243948... this was not even a rounding failure.

If we take this approach to your gaming situation, there are precise position parameters and time frames. The server can calculate precise if such an action could be possible - easy triangulating math. Why would you want to predict if it was possible or not if you just can calculate it? And AI probably does not even go the math route, because another one is "easier" (which does not mean more correct).

AI in games is good for things like animation prediction, can save resources while being much more natural than classic systems. But there is no benefit in using it for anti-cheat (except it may costs less at development, but it can damage the companies reputation in return).

Quoting: PlayingOnLinuxphone

Quoting: sarmadFor example, it should be able to detect that you are aiming precisely at a target behind a wall, or other similar things that are impossible without cheating.

That is a good example why classic algorithms are better than LLMs which are basically algorithmic lossy archive files. It is something like MP3 for all kind of data where the timestamp is the input (prompt, randomization number and so on). The important part is "lossy", because it loses information on the training process. This causes fail predictions as not being able to do correct maths. I just asked CGPT on duck.ai "What is the math result of 13/73²?" and the result was "13 / 5329 ≈ 0.002438 (rounded to 6 significant figures).", while the real numbers are 0.00243948... this was not even a rounding failure.

If we take this approach to your gaming situation, there are precise position parameters and time frames. The server can calculate precise if such an action could be possible - easy triangulating math. Why would you want to predict if it was possible or not if you just can calculate it? And AI probably does not even go the math route, because another one is "easier" (which does not mean more correct).

AI in games is good for things like animation prediction, can save resources while being much more natural than classic systems. But there is no benefit in using it for anti-cheat (except it may costs less at development, but it can damage the companies reputation in return).

You're making the same mistake I made.
[AI](https://en.wikipedia.org/wiki/Artificial_intelligence) isn't [machine learning](https://en.wikipedia.org/wiki/Machine_learning)
It's the capability of computational systems to preform tasks typically associated with human intelligence.
As proof I will provide an example from the gaming space.
Far for the AI boom we've been calling the behavioral logic behind npcs the AI.
That logic is coded in no machine learning involved, yet we call it AI.

We nowadays associate it with machine learning, because most modern AI companies use mostly machine learning to achieve their goals, but that's just an artifact of our time.

Last edited by LoudTechie on 30 Apr 2026 at 11:38 pm UTC

Can't wait for Re-DENUVO with double and triple the DRM appearing in games.

How long before this software suit requires age verification and constant camera feed to ensure you aren't doing something they don't like... like having fun...

Quoting: LoudTechieYou're making the same mistake I made.

Where I make a mistake? The context was clearly not about NPC behavior, but about the marketing term "AI", which I usually avoid.

Quoting: mr-victory

Quoting: doragasuUnfortunately there's a growing probability of Denuvo moving to kernel level DRM. And if that happens, it's game over for Linux users. Not a problem specifically for me, I don't buy games with Denuvo DRM, but for the platform itself, it might be a big hit.

Do you have a source? Hope this doesn't happen

No source, that's just me guessing. But with Denuvo pricing reported as $25K per game per month + 50 cents per game sold, I am sure they are doing all they can to stay relevant, it's big money in here. And given their protection seems completely defeated, I suspect going to kernel level might be the next step.

Quoting: PlayingOnLinuxphone

Quoting: LoudTechieYou're making the same mistake I made.

Where I make a mistake? The context was clearly not about NPC behavior, but about the marketing term "AI", which I usually avoid.

The context was AI based anti-cheat.
AI based anti-cheat is also older than the prolifiration of LLM's and the implicit connection between machine learning and AI.
Minecraft anti-cheat is the most known example of AI based anti-cheat and precisely as algorithmic and deterministic as you're proposing.
It just emulates human like analysis by looking at cheating incentives and realistic human behavior.

Quoting: PlayingOnLinuxphone

Quoting: sarmadFor example, it should be able to detect that you are aiming precisely at a target behind a wall, or other similar things that are impossible without cheating.

That is a good example why classic algorithms are better than LLMs which are basically algorithmic lossy archive files. It is something like MP3 for all kind of data where the timestamp is the input (prompt, randomization number and so on). The important part is "lossy", because it loses information on the training process. This causes fail predictions as not being able to do correct maths. I just asked CGPT on duck.ai "What is the math result of 13/73²?" and the result was "13 / 5329 ≈ 0.002438 (rounded to 6 significant figures).", while the real numbers are 0.00243948... this was not even a rounding failure.

If we take this approach to your gaming situation, there are precise position parameters and time frames. The server can calculate precise if such an action could be possible - easy triangulating math. Why would you want to predict if it was possible or not if you just can calculate it? And AI probably does not even go the math route, because another one is "easier" (which does not mean more correct).

AI in games is good for things like animation prediction, can save resources while being much more natural than classic systems. But there is no benefit in using it for anti-cheat (except it may costs less at development, but it can damage the companies reputation in return).

Yes, you should not depend on AI for doing math at all, but math alone cannot easily determine cheating; you need math + patterns. In other words, you need math to convert the data into a form that can then be consumed by AI. For example, assume you are aiming at someone and moving sideways at the same time, then you go behind an obstacle and you continue aiming at the same spot which remains precise then you continue moving until you are away from the obstacle and you shoot immediately and get a perfect shot. This is not cheating because you had an initial line of site that you used to aim, but doing math alone will determine that you are cheating because you are aiming behind a wall. On the other hand, if the subject moves while you are behind the wall and you successfully follow its movement while still being behind the wall then that's likely cheating. I say likely because this can very well be just a coincidence or it could be experience like knowing the direction of the subject and predicting it's movement. But if this happens all the time with high precision then the person is more likely cheating. So, the proper solution is to use math to convert the data into high level data, then feed that to AI for a smart cheat detection.

@LoudTechie: I don't discuss such meta nonsense with you. I don't even understand why you are arguing here something that is clearly out of context. So I did not make a mistake, you are just looking for a failure that does not exist in the underlying context. All neuronal networks are just bad in this job, no matter if LLM in specific or not. And "AI" that is no neuronal network (even if it is a smart algorithm) is no AI or not relevant to prevent cheating, because the term is meant to be connected as "artificial animal/human being" like NPCs (behavior simulation) or neuronal networks (algorithmic brain simulation).

Quoting: sarmadyou need math + patterns.

Patterns are nothing else than pure math. For simple things neuronal networks can be better indeed. But with growing complexity the failure rate is also growing significant. And games are just too complex to work well (you don't want false positives). So it is better to stay at simple rules and some games proofed that these are good enough for a cheating free experience. Your example is something I am pretty sure I could realize in my own game with classic algorithms. Cheaters that do such obvious things are even more easily to filter using kill-death-rate with a value of 10+:1 (because they are stupid enough not to stay in the range of best legit players).

Moreover classic data analysis can also handle complex data structures (and especially patterns) easily. That's the way search algorithms, social media profiles and a lot of other stuff was done before neuronal networks became an useful thing.

But let's say there is a way using neuronal networks in a good way to prevent or ban cheaters while the failure rate is 0.0%. Then it becomes accepted and lazy companies build in these tools in a lazy way and so the failure rate is not 0.0% any longer. So even in the theoretical best case scenario it is better to not open the door. I don't want to have the fear that AI can always take away my progress by auto-banning when I was playing totally fair (which I always do, because cheating takes away the fun of challenges).

And again, why you would want to implement such tool, if it brings no benefit over classic algorithms, but potential risks? The only point is "costs less money".

As I understand it, some games (such as Arc Raiders?) will group players with similar skill levels, which seems like a good way to limit frustration. Rather than banning cheaters, or someone who *looks* like a cheater, just make them play with other cheaters. If it's a false positive, it may be annoying, but way better than being banned?

Edit: I guess this is a bit off-topic, since the article is about DRM and online checks. But the conversation has drifted a bit into anti-cheat...

Last edited by Phlebiac on 2 May 2026 at 5:07 am UTC