Why games break

You’re deep in the flow state, it’s the final play of the game, you see the enemy and they have their back to you, this is it, one shot and then the euphoria of the win. Boom, you’re dead - game over. The kill cam shows what you already knew. That play was impossible, the other player wasn’t better than you. They were cheating.

Nobody likes to lose, but there’s nothing worse than losing to a cheat. It’s not just a loss, it’s a betrayal. You signed up for a fair fight. You put in the hours, learned the mechanics, developed the skills. And someone else decided that the rules didn’t apply to them.

That’s the reason that Anybrain exists.

That feeling, the one that makes you want to throw your controller at the wall, press alt-F4, or maybe even uninstall the game, that’s not a security problem. It’s a trust problem. And when enough players feel it, the whole ecosystem starts to crumble.

Studios know the surface-level stats: cheating drives churn, reduces retention, floods support queues with complaints and false reports and angry Reddit threads. But the deeper damage is harder to quantify. It’s the slow degradation of the community that makes multiplayer games worth playing in the first place. The unspoken agreement that says: we’re all here to compete for real.

When cheating runs rampant, it doesn’t just ruin a match. It poisons the well.

And the well is bigger than most people realise. Recent industry research estimates the global cheating ecosystem, cheat subscriptions, boosting services, account sales, currency farming; the whole shadow economy sits somewhere around a median estimate of $8.5 billion. To put that in perspective: that’s larger than the entire esports industry. A single Chinese cheat developer, busted in 2021, had made $77 million selling hacks for just one game. This isn’t just a nuisance, it’s a whole underground industry of its own.

“The whole shadow economy sits somewhere around a median estimate of $8.5 billion. To put that in perspective: that’s larger than the entire esports industry.”

Why we cheat

Before we talk about solutions, we need to understand the problem. And that means understanding cheaters as people, not just accounts to be penalized.

You can ban people but that doesn’t always stop people from cheating, to be effective it’s also good to understand why people cheat. That sounds obvious, but it’s actually a bit nuanced in the industry. Most anti-cheat companies talk about cheaters like they’re malware; fraudsters, threats to the ecosystem, a problem to be detected and eliminated. And fair enough, that is the job. But somewhere along the way, the conversation stopped being about players and started being about signatures, kernel access, and detection rates. At Anybrain, we’ve been building tech around HCI (Human computer interaction) and security since 2016. And honestly? We also fall into that trap. It’s easy to get so deep into the security layer that you forget there are actual humans on both sides. The players getting their games ruined, and yes, the people doing the ruining.

Zooming out, the uncomfortable truth is simply that cheating works. That’s why there’s a ghost industry. This isn’t even a game problem, it’s deeper than that. The short version is that cheats give our brains the shortcut to the reward without the grind. And in a world where games and tech are increasingly designed around progression, prestige, and status, that shortcut is genuinely tempting. The thrill isn’t always in the learning. Sometimes it’s just in winning, and in feeling powerful.

Games are social systems, but they’re also fantasies. We play them to escape, to feel powerful, to be someone we’re not in everyday life.

When a game sells you a power fantasy and then asks you to grind for hundreds of hours to achieve it, is it really surprising that some players look for the easy path?

Our brains love shortcuts and in some ways, cheaters are just following their natural wiring.

That doesn’t make it okay. But it does mean this isn’t a simple story of good players versus bad players. The motivations are ultimately human: the desire to win, to matter, to flex status in a world where you can control the outcome. Of course, there’s some cheaters who are malicious griefers who get off on ruining other people’s fun. Some are bored and curious. Some just want to see what happens when they break the rules. And some are so deep in rationalising their behavior, that they’ve convinced themselves everyone else is cheating too, so why shouldn’t they?

Not all cheating is created equal, either. Some are accepted and even lauded, speedrunners, for instance, exploit bugs and glitches and have built entire careers and communities around it. Modders push boundaries that eventually become official features. Game developers have god mode access to their own creations, in a way the original cheat code. The line between exploit and innovation has always been quite blurry.

None of that means cheaters should roam free and ruin the experience for everyone else. But treating every cheater as identical: same motivation, same intent, same punishment, ignores the complexity of the situation. And if we don’t understand the problem clearly, we’re going to keep building incomplete solutions. Just to be clear, we don’t have an answer to this either, but do believe that understanding the problem helps to better combat it.

The arms race

The games industry has been fighting cheaters for decades. And for most of that time, the approach has been the same: find the cheat software, block it, ban the player.

Traditional anti-cheat works like an antivirus software. It maintains a database of known threats: signatures, file hashes, memory patterns and scans for matches. When it finds one, it issues a flag or a ban. This worked for a time but things are changing. The problem is that cheat developers update their tools constantly. Every detection triggers a new version. Every patch gets reverse-engineered. The defender is always one step behind, reacting to what already happened instead of anticipating what’s coming.

This is the cat-and-mouse game the industry has been stuck with since its conception. And honestly? It’s exhausting for everyone involved.

Because of that some anti-cheat solutions go deeper; kernel-level access, rootkit-style monitoring, aggressive system scanning. These catch more cheats, but they come with real issues like privacy concerns, performance overhead, and false positives that ban innocent players with unusual hardware setups, VPNs, or just plain bad luck. Every wrongful ban is a support ticket, a refund request, a frustrated player who tells their friends, and now pretty much the whole world that the game punished them for nothing. The collateral damage adds up.

Meanwhile, the most sophisticated cheats, the ones built by people who actually know what they’re doing, the $200-a-month “undetectable” subscriptions with automatic updates are monetarily incentivised to find ways around even the most invasive detection. That’s people’s genuine livelihood on the line. So this arms race never ends. It just escalates.

There’s a structural reason for this - signature-based detection is fundamentally reactive. You can only block what you’ve already seen and in a world where new cheats appear daily, where cheat-as-a-service subscriptions offer detection guarantees and same-day patches, that’s a losing position.

Our reasoning is that the whole framing is wrong, we’ve been looking at the problem all wrong.

Our reframe was: what if we looked at the player instead?

Every player has a fingerprint. Not a literal one but a behavioural one; the way you move your mouse, the rhythm of your keystrokes, the micro-hesitations before you take a shot, the subtle differences in how your gameplay shifts, etc.

These patterns are as unique as a signature. And they are almost impossible to fake at scale.

This is the insight that changes everything: a cheat can spoof a file, but it cannot replicate the ten thousand micro-decisions a human makes every minute of play. Think about it, what does an aimbot actually do? It removes hesitation. It eliminates overcorrection. It produces mechanical precision where there should be human imperfection, that wobble, the delay, the slight overshoot and recovery that real movements always have. And that imperfection, is one of the things that makes genuine play identifiable.

The tell isn’t the cheat software. The tell is the absence of humanity in the inputs. When you look at behaviour instead of code, you’re not asking “does this player have cheat software installed?”

You’re asking “does this player move like a human?” Those are different questions with very different implications. And the second one is much, much harder to fool.

It works across the board, too. A smurf account plays differently than the player’s main, the skill is there but the behavioural fingerprint doesn’t match the account history. A boosted account plays differently than its owner. A bot farms with repetitive consistency that no human can match over hours of play. Hardware modifications like Cronus or XIM alter the input signal in specific, learnable ways that separate them from a genuine human control.

The fingerprint doesn’t lie, even when the credentials do.

Redefining the game we play

The question for studios has never really been “how do we catch cheaters?” It’s always been “how do we protect the experiences we’ve built?”

Those are different questions with different answers. One leads you to a ban list and an endless game of whack-a-mole. The other leads you to understand your players - who they are, how they move, when they’re thriving and when they’re about to leave.

That’s the game worth playing. And it starts with a simple shift in perspective.

Stop looking at the software. Start looking at the human.

The same creativity that makes games fascinating is exactly what makes them exploitable. We’ve spent enough time chasing code. It’s time to understand behaviour.

Traditional anti-cheat looks at software.

Anybrain looks at behaviour.

Cheats can hide in code.

But behaviour doesn’t lie.

The State of Cheating in 2026 was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Imagine you’re playing an RPG. Your character, let’s name him Nuno, gets to a crossroads, and you have two possible roads to follow: path A, a dark forest or path B, a rocky mountain. You can only choose one, but you can explore each path before deciding. In classical computing, Nuno can only explore one path at a time. You could start by exploring path A first, gathering data, bosses fought, loot found, traps fallen, and return. Then you explore path B, gather more data, and compare. This works, but exploring one at a time is time-consuming, and resources are limited! And now imagine if it were 100 different paths, exploring each one sequentially would be a serious grind.

But, what if we are not in a classical environment? What if your character has a special ability called superposition, which allows them to exist in multiple realities at once? Instead of exploring one path at a time, your character “splits” across all paths simultaneously. Now, with this ability, we can gather all the data we need in a fraction of the time! Also, in this version, you have a travel companion, named Carlos, and both Nuno and Carlos are linked. If Nuno finds a healing potion recipe, Carlos will also benefit from it. If Carlos falls into a trap, Nuno will adjust the course without even having to see Carlos in the trap. This is entanglement, in the state that the duo shares knowledge instantly, which improves the decision-making of the duo.

This teamwork means that they are evaluating all paths in parallel, like exploring every single quest of a storyline at the same time. This not only makes the decision process much faster but also more informed. So, now you’re at a point where Nuno’s and Carlos’ experiences interact and interfere with each other. Where successful paths, with high rewards, reinforce each other, and low-rewards or dead ends cancel each other out. In the end, when Nuno makes a final choice, it won’t be random; you won’t forget path number 75, or mix path number 2 with path number 6, all paths are weighted in this final choice, which is the most promising path.

Superposition, entanglement, parallelism and interference are some of the key phenomena that make quantum computing so different from classical computing. Not only is it faster, but smarter, more optimised for decision-making and better suited to scale with increasingly complex problems, like, what if we had 1000000 paths!

Of course, these phenomena have been simplified to fit within the RPG example. In reality, they involve complex mathematics and physics, which we won’t discuss here. The objective is just to give a feel for how differently quantum computing works from classical computing.

So, how does this matter in data science? Let’s think of every decision our RPG has to make as optimising models, searching for patterns, choosing model features, and many others. We can see how beneficial quantum computing would be for data scientists, who often have to deal with massive and complex problems. And, let’s be honest, classical tools struggle with high-complexity problems. This is where quantum computing shows promise, enhancing our ‘toolbox’ for the kinds of data challenges that are becoming too big for classical methods.

Treasure Hunting with Quantum Tricks

Let’s see this advantage through a practical example, still using our gaming analogy.

Imagine we have 32 paths, and only one of them hides a treasure. The rest are empty. This is a classic search problem, and to solve it, we can take two strategies: a classical brute force search, or a quantum search, using Grover’s Algorithm.

The classical Brute Force

In the classical approach, we search each path one at a time. In the worst-case scenario, there’s a possibility that we have to search all paths. This means we might have to check all 32 paths to find the path we want. So, for N possible paths, this strategy takes O(N) time.

Quantum Grover’s Algorithm

On the other hand, Grover’s algorithm can find the correct path in only √N steps! For 32 paths, this is about ~6 steps. Therefore, for N possible paths, it would only take O(√N).

To better understand the logic behind Grover’s search, let’s simplify the problem: suppose we know the exact path code we’re looking for, we just need to search for it within a list of possible paths? Imagine each path name is defined as a unique combination of 1’s and 0’s, like ‘01001’ — in quantum computing, such binary strings correspond to quantum states, represented in Dirac notation, just like |01001⟩.
For the brute-force approach, it is very straightforward, as the name suggests. We check each path; if it is what we want, it stops!

But, for Grover’s algorithm, it is not that simple. Remember reading about superposition, interference and entanglement in the intro? Well, that is what makes it fundamentally quantum!

We start with creating a superposition of all possible states, which means the quantum system is in a mix of all N paths at once (something only quantum computers can really do).

In the next step, we create a function called oracle, which is a quantum function that marks the path we want by flipping the sign of the state (our path name). For example, in classical terms, it’s like a function f(x) that returns 1 if x is what we are looking for, and 0 if otherwise. In the quantum version, we have:

This function sets up interference for the next phase, called the diffuser. In this step, it uses interference to amplify the probability of the states that correspond to the wanted solution, and fades the ones that do not.
It then proceeds to repeat the previous steps, for about N times. By repeating, it increases the probability of the correct path. Finally, the quantum system is measured, and it returns the solution state, with a high probability of being the path we were looking for!

from qiskit import QuantumCircuit, transpile
from qiskit_aer import Aer
from math import floor, pi, sqrt
import time
import matplotlib.pyplot as plt

def diffuser(n):
  # This function creates a diffuser gate for Grover's algorithm
  # This gate boosts the probability of measuring the correct path

  diff = QuantumCircuit(n) # Create a quantum circuit with n qubits
  # Perform some quantum gate actions that make up the diffuser
  diff.h(range(n))
  diff.x(range(n))
  diff.h(n-1)
  diff.mcx(list(range(n-1)), n-1)
  diff.h(n-1)
  diff.x(range(n))
  diff.h(range(n))
return diff

def phase_oracle(n, target_index):
  # This function creates the oracle, the gate that marks the correct answer
  # It flips the sign(phase) of the target state
  # The marking helps the diffuser know which answer to amplify.

  oracle = QuantumCircuit(n)
  binary = format(target_index, f'0{n}b') # Converts the target to binary (e.g., '111')
  for i, bit in enumerate(binary):
    if bit == '0':
      oracle.x(i) # Apply a flip
  oracle.mcx(list(range(n-1)), n-1) # Marks the target
  for i, bit in enumerate(binary):
    if bit == '0':
      oracle.x(i) # Undo the flips
return oracle

max_qubits = 20 # Set maximum number of qubits, in this case to run the Grover's algorithm for 2 to 20 qubits (search space)
backend = Aer.get_backend('qasm_simulator') # Get a virtual quantum computer (a simulator)
brute_steps = [] # Array to save the number of steps with the brute force approach
grover_steps = [] # Array to save the number of steps with Grover's algorithm
simulation_times = [] # Array to save how long the quantum simulation took

# For each number of qubits from 2 to max:
for n in range(2, max_qubits + 1):
  N = 2**n # Total number of possible paths
  target = N - 1 # Set the target as the last item ( to test the worst case scenario)
  num_iterations = floor((pi / 4) * sqrt(N)) # Calculate number of Grover steps
  
  # Create a quantum circuit
  qc = QuantumCircuit(n)
  qc.h(range(n)) # Initialise in superposition (put the system into a state of equal possibility)
  
  oracle = phase_oracle(n, target) # The gate that marks the correct answer
  diff = diffuser(n) # The gate that amplifies it
  
  # Repeat the process
  for _ in range(num_iterations):
    qc.append(oracle.to_gate(), range(n))
    qc.append(diff.to_gate(), range(n))
  qc.measure_all() # Measure the result

  # Simulate and time how long it takes
  start = time.time()
  compiled = transpile(qc, backend)
  result = backend.run(compiled, shots=1024).result()
  end = time.time()

  counts = result.get_counts() # Count how often each answer came up
  simulation_times.append(end - start)
  brute_steps.append(N)
  grover_steps.append(num_iterations)

# Draw a graph showing how many steps each method took
plt.figure(figsize=(12, 6))
plt.plot(range(2, max_qubits+1), brute_steps, label="Brute-force steps (O(N))", marker='o')
plt.plot(range(2, max_qubits+1), grover_steps, label="Grover steps (O(√N))", marker='s')
plt.xlabel("Number of Qubits")
plt.ylabel("Steps")
plt.legend()
plt.grid(True)
plt.tight_layout()
plt.show()

We’ve seen the basic idea behind Grover’s algorithm, but how does it perform as the number of paths grows?

Let’s look at the performance difference between the classical and quantum approaches mentioned above. In the plot below, we compare the number of steps each method takes as the number of ‘paths’ increases.

The blue line represents the brute-force performance, and the orange line represents the Grover’s algorithm performance. The blue line grows exponentially with the number of paths; on the other hand, the orange line grows much more slowly! At 220 paths, the brute-force needs over 1 million steps, as for Grover’s algorithm only needed about 1000 steps.

Quantum Advantage: What’s Real, What’s Next?

This is one of the examples where we can see the quantum advantage in action! Other algorithms play the same advantage, in different fields. For example, in cryptography and data security, we have Shor’s algorithm that can factor large numbers exponentially faster than popular classical algorithms.

And what about data science? Quantum computing holds a promising future here! Many problems in data science involve dealing with big datasets, finding patterns in high-dimensional spaces, and optimising models. All of these are tasks where we can see that quantum algorithms could provide an edge.

Not only can quantum computers process big amounts of data much faster than classical computers, but we also have Quantum Machine Learning algorithms, like Quantum Principal Component Analysis and Quantum Neural Networks. We also have Quantum Optimisation Algorithms that are used to solve optimisation problems, etc.

Even though we are in early stages, we can see the potential!

Of course, all of this comes with challenges. One of the biggest challenges is hardware. Current quantum computers are still limited and prone to errors, which restricts the complexity of problems they can process.

We do have advanced quantum computers developed or being developed by IBM, Google, among others. For example, IBM Quantum Eagle has 127 qubits — fundamental units of quantum information, similar to classical bits but with a very different behaviour. These qubits are extremely fragile and easily disrupted by external factors and other quantum phenomena, causing errors to accumulate during computation.

And while 127 bits may sound impressive, it is still quite limited for most quantum algorithms! To outperform classical systems on meaningful problems, we would likely need thousands to millions of qubits.

Beyond hardware, there’s also the challenge of developing the algorithms. Many quantum algorithms still remain on a theoretical basis. Developing them is non-trivial: it requires new ways of thinking about problems, new mathematical tools, and a whole new, unfamiliar logic!

In addition, quantum algorithms don’t always perform better than classical ones. They often only show an advantage when the data is very large and complex — that’s where they really shine — something current systems can’t fully support yet.

Finally, there’s a resource gap. Only a small group of researchers has access to actual quantum machines. Most people rely on simulators, which don’t reflect the full potential of a quantum computer.

A Final Challenge for the Reader

We’ve seen how quantum computing opens up a world of new computational possibilities. Now, let’s take a moment and imagine this in a gaming context.

• What happens when cheat developers have access to quantum-powered tools?
• What if they create a cheat software that could simulate thousands of gameplay outcomes in parallel, and could adapt in real-time?
• Could quantum machine learning help mimic human behaviour better than ever before?

Could this even become reality? This may be thinking too much ahead, but we need to stay ahead of the curve; the race is on! So we leave you with this challenge:

How would you use quantum principles to generate the next generation of cheating software?

Bibliography

Nielsen, M. A., & Chuang, I. L. (2010). Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge University Press.

IBM Quantum. IBM Quantum Experience. Retrieved from https://quantum-computing.ibm.com

Qiskit Textbook. Grover’s Algorithm. Qiskit by IBM. Retrieved from https://qiskit.org/textbook/ch-algorithms/grover.html

Preskill, J. (2018). Quantum Computing in the NISQ era and beyond. Quantum, 2, 79. https://doi.org/10.22331/q-2018-08-06-79

Montanaro, A. (2016). Quantum algorithms: an overview. NPJ Quantum Information, 2, 15023. https://doi.org/10.1038/npjqi.2015.23

Quantum Quest: An RPG Tale of Superposition and Entanglement was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

In this article, we give the floor to the first number of a series of texts about data science. Data Science is essential to Anybrain’s mission of protecting all types of games on all platforms. The data we collect from human-computer interaction would be meaningless without it, and we couldn’t advise studios on the state of their games effectively.

Our data science team drives innovation in the gaming world — ensuring fair play, detecting fraud, and pioneering new ways to enhance player experiences. We recently conducted a roundtable interview with every team member, asking the same questions to learn about their personal journeys, roles, daily routines, challenges, future projects, and visions for the future of gaming.

Below, we’ve organized their insights by question, offering you a comprehensive look into the heartbeat of our team.

1. Tell Us About Yourself, Your Academic Background, and How You Joined Anybrain

Gustavo Gomes
Gustavo introduces himself with humor and candor: “I am getting old — I’ll probably be 30 already when this is published.” With a background in software engineering and studies in AI and computer graphics, he originally dropped his thesis (which didn’t feel meaningful) in favor of a gap year chasing a dream of being a professional Dota 2 player. Five years on, he’s back to finishing his thesis — this time on fraud detection using anomaly detection techniques in gaming.

Carlos Gomes
Carlos, a Computer Engineering graduate from the University of Minho, shares that his love for technology is deeply personal — enabled by his cochlear implant. A lifelong passion for gaming led him to Anybrain, where a recommendation from a colleague turned his summer internship into a full-time role. His journey is a testament to curiosity and the transformative power of technology.

Pedro Duarte
Pedro, born in Braga and holding a Master’s in Computer Engineering specializing in Intelligent Systems and Language Processing, began his career at Anybrain with a master’s thesis on esports performance. His work laid the foundation for Anybrain’s pivot toward the gaming sector, setting him on a path that now has him leading the Data Science team.

Nuno Silva
Nuno’s story begins in his hometown of Braga, where he completed an Integrated Masters in Computer Engineering at the University of Minho. A summer internship introduced him to AI/ML, and his master’s dissertation — supervised by André Pimenta (Anybrain’s CEO and co-founder) — sealed his connection with Anybrain. His passion for technology and gaming shines through in every project he undertakes.

Ana Cruz
Ana brings a unique blend of academic rigor and gaming passion. With a degree and master’s in Applied Analysis and Computation in Mathematics, she found Anybrain while exploring Portuguese startups. Excited by the opportunity to merge her analytical skills with her love for gaming, Ana joined the team to help fight cheating and ensure fair play.

2. What Is Your Specific Role Within the Team?

Gustavo Gomes
Gustavo is charged with discovering and developing new fraud detection solutions. Beyond crafting innovative models, he also monitors existing detections, performs ad hoc data quality analyses, and builds data pipelines as new challenges arise.

Carlos Gomes
Carlos focuses on creating a seamless data pipeline — from initial data preparation to training and optimizing models, and finally deploying them to catch cheaters. His role ensures that every stage of model development is executed with precision.

Pedro Duarte
As the Team Leader, Pedro coordinates the entire process: from data collection (analyzing gameplay inputs) to feature engineering, model evaluation, and even customer support. His role encompasses managing the infrastructure and guiding the team through technical challenges, ensuring that every detection meets high-quality standards.

Nuno Silva
Nuno works as a Data Scientist primarily focused on developing AI/ML models that enhance fraud detection. His contributions are key to advancing the technology behind Anybrain’s anti-cheat efforts.

Ana Cruz
Ana is part of the Data Analysis team. Her daily tasks involve ingesting and scrutinizing vast amounts of gameplay data using tools like Python, pandas, and NumPy. By applying statistical methods and machine learning algorithms, she transforms raw data into actionable insights that help identify anomalous behavior.

3. How Is a Normal Work Day for You?

Gustavo Gomes
Gustavo’s day varies considerably. Some days are spent coding — switching between R and Python for quick analyses — while other days are devoted to research and strategic thinking. He emphasizes the importance of “zooming out” to foster new ideas.

Carlos Gomes
Carlos’s routine starts with detailed data analysis before moving on to model training. He fine-tunes hyperparameters and employs interpretability tools like SHAP and LIME. Between exploring new techniques and keeping up with the latest in gaming (Silksong news included!), his day is a dynamic mix of technical and creative work.

Pedro Duarte
Pedro begins his day with a gym session, then dives into task reviews and prioritization. His schedule includes monitoring customer support, processing metrics in R/C++, leading team meetings, and strategic discussions with leadership. Evenings are reserved for research and catching up on the latest trends, balancing a demanding workload with continuous learning.

Nuno Silva
For Nuno, mornings are often dedicated to reading articles on Medium to stay updated on industry trends. He follows a structured routine that includes a gym break at lunch, intensive data preparation sessions, and model development. In his downtime, he relaxes with gaming or language classes.

Ana Cruz
Ana’s day starts by identifying a particular cheat or behavior to analyze. She then dives into research — exploring academic articles and documentation — to select a suitable method. Once she applies her chosen approach to the data, she iterates and refines her models before wrapping up the day by unwinding with some gaming, gaining firsthand insights into the player experience.

4. What Were the Biggest Problems/Barriers You Surpassed at Anybrain?

Gustavo Gomes
Gustavo highlights the inherent challenges of working with machine learning in a small company. Balancing extremely large or tiny data volumes along with computing constraints demands a great deal of tenacity — a quality he and his team have honed over time.

Carlos Gomes
Carlos reflects on both technical and personal hurdles. Initially, he had to adjust to massive datasets far beyond those on Kaggle, and he faced challenges in communication due to his cochlear implant. These experiences pushed him to grow both professionally and personally, ensuring that models accurately differentiate between normal and fraudulent behaviors.

Pedro Duarte
Pedro shares that his journey involved managing a demanding work schedule and transitioning from a one-person data role to leading an expanding team. He had to learn on the fly in an unstructured environment, particularly balancing the accuracy of fraud detection (minimizing false positives) with robust model development. Additionally, mastering the tools and infrastructure required continuous adaptation.

Nuno Silva
For Nuno, the primary challenge has been mastering organization and communication. Learning to break down and distribute his tasks effectively throughout the week has been key to navigating the complex landscape of AI/ML projects.

Ana Cruz
Ana’s biggest hurdle was transitioning from a theoretical academic background to delivering practical, actionable insights in a fast-paced environment. She also had to significantly enhance her programming skills to handle the scale and complexity of gaming data — a challenge she has met head-on.

5. Which Future Projects Do You Have in Mind?

Gustavo Gomes
Gustavo envisions a project aimed at detecting any new cheating tricks as soon as they emerge. He believes that, although it will take time to perfect, such an initiative could become a cornerstone in creating the ultimate AI-based detection system.

Carlos Gomes
Carlos is keen on exploring the deployment side of things — specifically, diving into MLOps. His future goals include learning and applying tools like Kubeflow, Kafka, Kubernetes, and Docker to ensure models are efficiently deployed and continuously updated in production environments.

Pedro Duarte
Pedro plans to expand the capabilities of internal tools already in use at Anybrain, such as the experiment tracking and fraud visualization platforms. He’s also excited about collaborating with customers with the goal of an exciting new feature (that we will share in the future), which would significantly enhance detection capabilities and tailor solutions to specific gaming environments.

Nuno Silva
Nuno doesn’t have a specific project lined up yet but is eager to deepen his knowledge of emerging technologies. He’s particularly interested in learning Spark, Kubeflow, and experimenting with Transformers to push the boundaries of what AI can do in gaming.

Ana Cruz
Ana is intrigued by the potential of quantum computing. Although still an emerging field, she sees it as a promising avenue to process and analyze data at unprecedented speeds, potentially revolutionizing cheat detection and pattern recognition in gaming.

6. How Will Data Science (or AI) Shape the Future of Gaming?

Gustavo Gomes
Gustavo believes that AI and data science will revolutionize gaming — not only by enhancing game features and competitive fairness but also by opening doors to new gameplay experiences. Imagine fair matchmaking, smart in-game assistants, adaptive economies, and even NPCs that act as integral game mechanics.

Carlos Gomes
Carlos sees data science as the key to crafting personalized, adaptive gaming experiences. By leveraging AI to adjust difficulty levels, rewards, and overall game complexity in real time, games can become more immersive and engaging — tailoring experiences to each player’s unique skill and preference.

Pedro Duarte
Pedro envisions AI as a transformative force across the industry. He predicts enhanced gameplay through adaptive challenges, more intelligent NPCs, and even streamlined content creation based on existing narratives. Furthermore, improved fraud detection ensures a secure, fair gaming environment — a balance of creativity and rigorous security.

Nuno Silva
Nuno echoes the sentiment of dynamic adaptation. He believes that as data science matures, games will continuously evolve to meet player preferences, making experiences more immersive through automatic adjustments in difficulty, rewards, and narrative elements.

Ana Cruz
Ana emphasizes that data science not only strengthens anti-cheat systems by detecting subtle behaviors but also enhances the overall gaming experience. From improved matchmaking to personalized gameplay and content development, data-driven insights will create secure environments that are engaging, and uniquely tailored to each player.

Final Thoughts

The insights from our team reveal a shared passion for gaming and technology — a blend that’s driving the future of fair play and immersive experiences. From innovative fraud detection methods to cutting-edge fields like quantum computing, the Anybrain data science team is committed to pushing boundaries and redefining what’s possible in gaming.

Other texts of this series will come out this year as we aim to showcase different aspects of our work and how our technology is shaping the present and future of video game protection.

Data Science for Gaming: the Anybrain’s team was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

What a year 2024 has been! It was challenging but also incredibly rewarding for the Anybrain team. In 2024, we brought some of the biggest game releases onto our platform, doubling our player base for more than 40M and tripling the number of gameplay hours analyzed. Additionally, we expanded Anybrain to new platforms and supported a wider variety of games — all while delivering massive value.

One of the highlights was working closely with our amazing partners and
customers. This teamwork helped us tackle some tough issues and really boosted what Anybrain can do. Thanks to these strong partnerships, we’re heading into 2025 with tons of enthusiasm and confidence in enhancing the gaming experience and building healthier communities in the video game industry.

So, what’s next for 2025? Even more commitment and passion!

We’ve made significant improvements over the past few months, enhancing our technology and expanding our capabilities to better serve the gaming community. And this is just the beginning of what we can achieve and deliver. Plus, we’re thrilled to share some of our secret projects finally and involve the community in our journey — there’s so much more to come that we can’t wait to reveal!

A huge thank you to all our partners, customers, and the awesome Anybrain team for your continuous support. Here’s to an even more innovative and successful 2025!

Looking Back at 2024 and Excited for 2025! was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

The video game industry is growing at an impressive pace, with new games being released every day, from innovative indie games to highly anticipated AAA titles releases from major studios.

One of the primary focuses of these games is the multiplayer modes, a cornerstone of modern games, that attract millions of competitive players eager to build their legacy.

Many games offer rewards to those players who reach the top tier of the game, it could be in-game benefits or even financial incentives.

These elements become even more significant in titles that have their own economies, such as free-to-play games, where rare items, skins, or virtual currencies can have real monetary value.

With the promise of great rewards and a fierce competition, an inevitable reality emerges, many players resort to cheats or prohibited methods to achieve the best results, and claim the best rewards that the game has to offer.

In this post, we’ll explore how AI anti-cheat systems and bans automation are transforming the way games combat cheating.

The Negative Impact of Cheats in Games

These actions affect both the player’s experience and the reputation of the studio behind the game. When no measures are taken or the best mechanisms are not implemented to combat cheating, games can quickly lose their player base, retain new users, and even face project shutdowns.

The negative image that was created can also compromise future projects from the studio and discourages potential investors.

The damages are not only related to reputation, in free-to-play games, that are highly dependent on their in-game economies, could see a great impact, where rare item lose their value, for example.

This often results in revenue loss for games that rely on microtransactions.

Solutions Against Cheats

There are several measures you can implement to mitigate this problem, the most common are:

• Server authority and encryption to protect game data and prevent external manipulation.
• Regular testing, including stress tests and debugging sessions, to identify and fix exploits, bugs, and glitches.
• Anti-cheat systems, such as Advanced AI solutions, like those offered by Anybrain, which analyzes and detects suspicious behavior in real-time with a level of precision higher than traditional methods.

However, merely detecting cheaters is not enough. It’s essential to impose effective and dissuasive sanctions on players who break the rules. These sanctions are often implemented in the form of bans.

Types of Bans

Bans can be applied in different ways, depending on the severity of the infraction and the game rules:

• Temporary bans for less severe infractions.
• Permanent bans reserved for more serious cases.
• Chat bans to block toxic behavior without banning players from the game.
• Account, IP, or hardware bans to prevent cheaters from easily returning.
• Shadow bans, isolating the cheaters from the legit players without their knowledge, minimizing their impact on the other players experiences.

Today, thousands of cheats are available in online forums, ranging from free tools to paid ones or with subscription services. Some are easily detectable, while others are more complex, making detection a challenge that only an advanced AI anti-cheat system can handle.

Free-to-play games face the additional challenge of players creating multiple accounts to continue cheating after being banned, and, if not resolved, can severely impact the community.

The first challenge starts here, even if cheaters are being detected, manually reviewing cases or applying bans is not a viable solution in games with thousands of players.

This is where bans automation becomes an essential part of your game.

What Is Bans Automation?

Bans automation is an automated system that applies sanctions to the players, it’s quick, scalable, and precise, without requiring constant human moderation. This system is particularly important in games with large player bases, where traditional moderation methods would be slow and costly.

In a game with over 10 million active users, ban automation could reduce manual moderation time by 80%.

Advantages of Bans Automation

By defining the scope of your rules and the scenarios in which they will be applied, including the type of ban, by adopting bans automation you can revolutionize how your game handles cheaters.

Here are the key benefits:

1. Efficiency
   Bans are applied automatically, freeing up moderators to focus on more important tasks such as community building or addressing critical issues.
2. Accuracy
   Rules are carefully defined and fully customizable, ensuring that bans are only applied to players who meet all parameters. This significantly reduces the risk of errors compared to manual bans.
3. Scalability
   In games with millions of players, manual ban application becomes impractical. Bans automation eliminates that limitation, allowing the system to operate seamlessly even during peak CCU hours.
4. Speed
   The system can apply bans in real-time, preventing cheaters from causing any harm or interference in the game. Alternatively, ban waves can be scheduled to remove multiple cheaters simultaneously, creating a strong deterrent effect.
5. Transparency and Control
   It’s automated, but the system allows you to review, update, or deactivate rules at any time. This ensures that you take complete control over decisions made by the system.

When well implemented, bans automation can effectively keep your game free of cheaters.

Addressing False Positives

Another challenge arises, we all aim to maintain a cheat-free game environment, but it’s important above all else, to avoid banning innocent players, the so-called false positives.

Although bans automation offers numerous advantages, addressing the issue of false positives is critical. This is one of the player’s biggest concerns, as any mistake could cause irreparable damage to the game and player reputation.

To mitigate this risk, we at Anybrain focus on maintaining a false positive rate below 0.001% for each detection. Additionally, the system allows moderators to manually review specific cases, ensuring that any uncertain situations can be thoroughly analyzed.

With bans automation, you can address potential false positives by applying a shadow ban to players with fewer fraud occurrences or minimal fraudulent activity.

This allows you to monitor their behavior over time, and if the number of frauds continues to increase, you can then apply a stricter ban.

Alternatively, if the player proves to be a false positive, you can revoke their ban.

This approach ensures proactive action on early frauds, helping you address potential issues while reducing the risk of excessive punishment.

Players Perception

There may be some player pushback against automated bans, feeling that the human touch in moderation has been lost. However, it’s essential to highlight that bans automation does not exclude manual moderation. Instead, it complements and enhances existing processes.

Games can also share aggregated data on ban statistics with the community to demonstrate the system’s reliability and fairness.

By clearly communicating the benefits and transparency of the system, studios can build trust within the community, reinforcing the message that the primary goal is to create a fair and competitive environment for everyone.

Anybrain Role in Combating Cheats

At Anybrain, beyond providing an AI anti-cheat capable of detecting the most complex cheats in real-time, we also offer tools such as:

• Bans automation, enabling automatic sanctions based on anti-cheat detections.
• Detailed analysis tools, allowing manual review of specific cases.
• Webhook integration, synchronizing information with the game system for better control and data analyzes.

Additionally, we provide continuous support to ensure that our system is always optimized and tailored to your games needs.

Conclusion

Bans automation is not just a tool, it’s a commitment to fairness and integrity in gaming.
Protecting your game means protecting your brand and your players.

Visit https://anybrain.gg/ to learn how Anybrain AI anti-cheat and bans automation can help you to detect and prevent cheats in online multiplayer games, how to increase your player retention and protect your game.

Your game, your rules.

Combating Cheating in Video Games with Bans Automation was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Part 2: Deployment Strategies

Turning Models into Real-World Solutions

Deploying machine learning (ML) models is crucial in turning your ideas into real-world solutions. While building the model itself is essential, making it work smoothly in a production environment comes with challenges. Your model must be reliable, scalable, and able to handle changing conditions without downtime.

This post will focus on the deployment phase — an essential part of the MLOps lifecycle. Choosing the right deployment strategy is not a one-size-fits-all decision. Your approach can significantly affect how well your system scales, how reliable it is, and how satisfied your users are. Whether you’re experimenting on a small scale or running large, enterprise-level models, it’s essential to understand the different deployment options.

This article builds on our previous post about getting ML models ready for production, where we covered the unique challenges of operationalizing machine learning. Today, we’ll dive into common model deployment strategies, their key features, benefits, and typical use cases.

Overview of Deployment Approaches

Selecting the right deployment strategy ensures a smooth transition from development to production. Below are the most common approaches, when to use them, and their pros and cons.

Shadow Deployment

In shadow deployment, we can run a new model alongside your current model without affecting users. The live model handles requests while the latest model works in the background, allowing you to test performance safely.

Pros:

• No impact on the user experience.
• It is easy to compare performance with the live model.
• Reduces risk by keeping the new model separate.

Cons:

• Resource-intensive, as two models run at once.
• No direct user feedback on the shadow model’s predictions.

Best for: Safely testing new models in real-world conditions without user exposure.

A/B Testing

In this strategy, we split user traffic between two models to compare performance. Users see one model or another, and you measure metrics like accuracy or engagement to decide which is better.

Pros:

• Get real user feedback.
• It is ideal for user-focused applications like e-commerce or recommendation systems.

Cons:

• Requires a large user base for meaningful results.
• Sensitive to traffic variations and data quality.

Best for: Directly comparing two models to find the best performance.

Multi-Armed Bandit (MAB)

Inspired by reinforcement learning, this approach directs more traffic to better-performing models over time, automatically optimizing which model gets the most users.

Pros:

• Adapts quickly and efficiently to improve performance.
• Less resource waste compared to simple A/B testing.

Cons:

• It is computationally expensive and requires advanced monitoring.

Best for: High-traffic scenarios, such as dynamic pricing, where rapid optimization matters.

Blue-Green Deployment

In a blue-green deployment, we use two environments: one (blue) runs the current model, while the other (green) hosts the new model. After testing the green environment, you switch traffic over with minimal downtime.

Pros:

• There is little to no downtime.
• Easy rollback if problems arise.

Cons:

• It can be costly due to duplicate infrastructure.
• Requires careful coordination between environments.

Best for: Systems where downtime is not acceptable.

Canary Deployment

In the canary deployment strategy, we introduce a new model to a small group of users. If it performs well, gradually increase the number of users who see it.

Pros:

• Real-world testing with low risk.
• Gradual rollout and easy rollback.

Cons:

• It needs strong monitoring to catch problems early.
• Slower rollout.

Best for: Situations where a careful, step-by-step release is essential, especially in consumer-facing apps.

Making the Right Choice

To deploy your machine learning models effectively, you must understand the various strategies available and when to apply them. Your choice will significantly influence your machine learning systems’ reliability, scalability, and overall performance in a production environment.

Consider the technical and business aspects when determining which deployment approach is best for your project. Each strategy has advantages and disadvantages, so the right choice will depend on model complexity, business objectives, and available resources.

Model Complexity: More complex models, such as those that rely on deep learning architectures or large-scale datasets, may benefit from strategies that allow safer testing before full rollout. For example, a shadow deployment can help validate complex models without disrupting the user experience, while a canary deployment can reveal performance issues at a manageable scale.

Business Requirements: Consider your tolerance for downtime or performance dips. If your application must remain constantly available, a blue-green deployment provides seamless transitions with minimal disruption. On the other hand, if your primary goal is to quickly gather user feedback, A/B testing or a multi-armed bandit approach might better align with your business objectives.

Resource Availability: Different strategies have different costs. Blue-green deployments, for instance, require parallel production environments, which increases infrastructure overhead. Shadow deployments, which run multiple models simultaneously, can be resource-heavy. Assess whether you have sufficient computing, storage, and budget before committing to a specific approach.

Risk Tolerance and Compliance: High-stakes applications — such as those in finance, healthcare, or critical infrastructure — often demand safer, more controlled rollout strategies. Shadow or canary deployments can mitigate risks by detecting issues early while preserving user trust and meeting regulatory requirements. If speed and innovation are top priorities, multi-armed bandit strategies can accelerate iteration but come with a higher operational risk.

By evaluating these factors, you can choose a deployment strategy that aligns with your technical capabilities, business objectives, and resource limitations. This thoughtful approach ensures that your selected method meets the requirements of your machine learning project and supports long-term success and growth.

Practical Guidance

Choosing the right deployment strategy is only the beginning. To achieve lasting success, it is crucial to implement effective practices for monitoring, maintenance, and continuous improvement.

Automate Wherever Possible: Implement Continuous Integration and Continuous Deployment (CI/CD) pipelines to streamline the model update process. Automated tests, performance checks, and validation steps can help catch issues early and maintain consistent quality.

Invest in Robust Monitoring and Alerting: Set up comprehensive monitoring to track key performance indicators like latency, error rates, and throughput. Real-time alerts can help your team quickly address problems as they arise, minimizing downtime and user impact. Logging and metrics tools can also provide insights into user behavior and model performance under different conditions.

Establish Clear Rollback and Recovery Procedures: No matter which strategy you choose, be prepared for the unexpected. Have a well-documented rollback plan that allows you to revert to a stable version if performance degrades quickly. A clear incident response process ensures smoother handling of unplanned outages or model failures.

Gather User Feedback and Iterate: Incorporate user feedback loops where possible. A/B testing and canary deployments can help you measure how changes affect user behavior for front-end applications. For back-end services, shadow deployments and monitoring tools can guide adjustments. Use these insights to refine models and enhance user satisfaction over time.

Start Small and Scale Up: To minimize risk and complexity, begin with limited rollouts. Test your chosen strategy on a subset of traffic or a single region before expanding to a full-scale deployment. Gradually scaling your approach helps validate assumptions, ensure stability, and identify scaling bottlenecks early.

Align Deployment Strategy with Business Goals: Regularly revisit your deployment approach to confirm it meets your organization’s evolving needs. As your traffic grows, your model’s complexity increases, or your business priorities shift, be ready to adjust your strategy. Flexibility ensures that your deployment methods continue to deliver the best possible results.

Aligning Deployment Strategies with Long-Term Success

Choosing the right deployment strategy is crucial in successfully operationalizing your machine learning models. Each approach — whether shadow testing, A/B testing, blue-green deployments, canary releases, or multi-armed bandits — has its strengths, limitations, and ideal use cases. You can select a method that best aligns with your organization’s goals by evaluating model complexity, business requirements, resource availability, and risk tolerance.

Once you have chosen a strategy, it’s essential to maintain strong operational practices such as effective automation, robust monitoring, and clear rollback procedures. Regular user feedback loops and iterative improvements help ensure that your deployment approach continues to support evolving business needs and technical demands.

However, there are still important questions that deserve deeper exploration:

• What metrics should be used to evaluate the performance of different deployment strategies?
• How can monitoring be effectively implemented during the various deployment strategies to ensure model performance and user experience?

These considerations will help refine your deployment workflows, ensuring that your models perform consistently and deliver value over the long term. In our next blog post, we’ll tackle these questions head-on, offering detailed guidance on metrics selection, monitoring best practices, and the essential tools and platforms that can streamline your ML deployment process. Stay tuned!

Productionizing Machine Learning Models was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

As gaming communities grow, so does the urgency of maintaining safe and respectful environments for players. Toxicity — including hate speech, harassment, and discrimination — has a profound impact on players’ well-being, engagement, and overall enjoyment. AI technology has emerged as a vital tool to detect, address, and reduce toxicity in real time, fostering inclusivity and safety in these spaces. With advances in AI, developers can deploy tools that promote healthier, more inclusive gaming environments while balancing player privacy and freedom of expression.

In this article, we will explore the many ways that AI can help moderate toxicity in gaming, some concerns that every AI company must have while building these types of tools, and finish exploring what the future holds for this particular issue.

Real-Time Detection: The Frontline of AI Moderation

One of the most transformative applications of AI in gaming is real-time moderation. Innovations like Modulate’s ToxMod monitor voice chat interactions as they happen. This is crucial in high-stakes or multiplayer games where interactions unfold rapidly. Tools like ToxMod analyze speech patterns and understand conversational context, helping distinguish friendly interactions from truly harmful language. This nuanced approach prevents unjust penalties, fostering authentic engagement while curtailing genuinely harmful interactions.

Adaptive Learning for Evolving Language and Use of Slang

AI-driven tools excel at learning from vast amounts of conversational data, enabling them to recognize evolving slang and detect varied forms of harassment or discrimination. Summa Linguae emphasizes that AI’s adaptability is crucial for identifying toxic language that traditional, keyword-based filters might overlook. By continuously updating their datasets, these tools ensure better accuracy and fewer false positives, creating a safer space for players without stifling expression.

Multi-Modal Moderation: Addressing Toxicity Across Different Media

AI’s ability to moderate multiple media — text, video, images, and voice — allows for comprehensive content control. Platforms like Checkstep integrate multi-modal AI moderation and are adaptable to various gaming environments and content formats. This multi-layered approach is particularly effective at reinforcing a platform’s unique community guidelines while remaining compliant with global standards. For example, a platform may permit competitive language within limits but curtail threats, harassment, or hate speech across all media.

Ethical Considerations: Privacy, Free Expression, and Bias

AI moderation introduces important ethical considerations. Privacy is a major concern, especially with real-time monitoring of voice chat or user behavior, where sensitive data may be collected. Dominic DaCosta’s research suggests that AI systems need clear, transparent policies on data handling to maintain user trust. Gaming platforms must address these ethical challenges by ensuring data protection and transparent moderation practices, including allowing players to contest moderation decisions.

Balancing user freedom with safety is another concern. Automated moderation risks unintended censorship of diverse viewpoints or cultural expressions, which can undermine community integrity. To maintain an inclusive space, DaCosta emphasizes that developers should employ culturally and linguistically diverse datasets. Such considerations enable the AI to recognize nuanced forms of toxicity — such as sarcasm or regional slang — without penalizing innocent interactions. Without these precautions, moderation systems risk alienating certain players or inadvertently scaling human biases that may exist in the data.

Overcoming Algorithmic Bias

One of the challenges with AI-driven moderation is algorithmic bias. Since algorithms rely on training data, they may unintentionally reinforce existing biases if datasets are unbalanced or unrepresentative. For example, tools trained predominantly on English-speaking datasets may inaccurately moderate non-English or culturally specific speech. This can result in biased actions, potentially marginalizing underrepresented player groups. Addressing these biases involves prioritizing fairness and accountability within AI design. Developers must regularly audit and adjust their algorithms, ensuring that moderation remains fair and inclusive for all community members.

Enhancing Transparency and User Trust

Transparency is key to user acceptance of AI moderation. By explaining how moderation works and enabling players to appeal decisions, gaming platforms can build greater trust. DaCosta’s research stresses that ethical AI use requires a transparent approach that empowers users and aligns with community values. This transparency, along with algorithm updates and community feedback, helps maintain moderation that adapts effectively to gaming cultures as they evolve.

The Future of AI-Driven Moderation in Gaming

AI-driven moderation holds transformative potential, but it requires careful implementation to balance user safety with freedom of expression. By addressing biases, enhancing linguistic and cultural awareness, and maintaining transparent policies, developers can foster healthy, inclusive communities. AI has the capacity not only to automate responses to toxicity but also to cultivate positive, diverse gaming spaces where players feel respected and valued.

For developers and platform operators, ethical AI practices will define the future of gaming. With continuous refinement and input from diverse community voices, AI systems will become increasingly adept at creating safer, more engaging, and inclusive gaming experiences. In an industry where community is central to success, prioritizing player safety and inclusivity will set leading platforms apart, establishing them as trusted spaces for enjoyment, growth, and connection. As gaming continues to grow, AI can play a big part in ensuring that gaming remains a positive, vibrant experience for all players.

Bibliography

DaCosta, Dominic. Enhancing Online Gaming Environments: The Role of AI in Moderating Toxic Behaviors. University of Virginia. 2024

AI’s Role in Moderating Toxic Behavior in Gaming was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

Of all the aspects of video game production, sound design is the one that takes my imagination further away. This happens because I know almost nothing about what happens in the magic sound designers and producers do in our beloved industry.

To better understand everything this world has to offer, I had the chance to talk with Pablo Schwilden Diaz, CTO at Demute, about his job and my curiosities about this topic.

So, let’s dive in.

First,

General Overview

What is sound design for video games? Does it sound different from other sound designs? (pun intended). In this first part of the interview I tried to make Pablo explain to me the skeleton of sound design for gaming, the ins and outs, and for the first question, I wanted to know about the fundamental aspects of it to create immersive experiences for players. Pablo said that “the most essential is to always try to look at things from the player’s perspective. It’s easy to get lost in the technicalities of game sound design and lose track of the bigger picture (…) To try and find the best experience, I always think back to the goals of video game audio such as Information, Immersion, and Interest. Any sound created has to fit at least one of those goals. Does the sound convey any information to the player, and if so does it convey the correct information? Does the sound help immerse the player in the experience or does it take them out of it? Does the sound create interest and curiosity in the player or does it bother and confuses the player? If a sound doesn’t work towards these goals, then it must be scrapped and started again.

So, here you have it folks, information, immersion, and interest! Sounds incredibly reasonable to me! I mean, who doesn’t recall that scare jump playing a horror game or the angelic music when you start to face another boss arena on a map or the music that slowly builds to keep you interested in a quest! Well, but if it wasn’t for Pablo I wouldn’t put the pieces together so it’s time to keep exploring this world.

Let’s go to,

Sound Effects and Foley

First things first, time to understand exactly what is the process behind the creation of sound for video games, does one start by the score, or by the ambient sounds? There is so much to do that I can’t possibly guess what is what. So, let the expert explain. Pablo says “The first step is to conceptualize the sound. That might sound obvious but many people forget the importance of this step and regret it later. This includes gathering some basic information such as: what the sound represents, what context will it play in, what’s the goal of the sound, etc. The more info we have here and the more decisions we can already make, the easier the rest of production will be.”

Of course! Prepare and plan first. How could I forget this part…

But it doesn’t stop there. The next steps are, as Pablo explains, “Then we wait. Seriously, it might sound weird, but more often than not, it is important for us to have a finished animation, VFX, or level to be able to create the best sound for it (…) Depending on the production and the amount of work, that waiting period is the perfect time to prepare material. Let’s say that we are working on a new creature, and we know from the concept phase that we will need a lot of material from a specific animal (let’s say a llama). While the modelers and animators work on that creature, we can go look for llama recordings or create them ourselves. This way, once we start sound designing, we already have all the necessary materials and are ready to work!

Once we have prepared this material and the necessary content is completed, we create the sounds needed. This will usually be done inside a DAW (Digital Audio Workstation), such as Reaper, by combining, processing, and editing different audio sources (recordings, synthesizers, etc) into one sound. When we are satisfied with the sounds we built, we export them as individual files following the format conventions decided during the concept phase.

After that comes the implementation phase. This will vary greatly depending on the tools and game, but the idea generally stays the same. Implementation is the step where we will define how a sound should play. Things like volume, reverb, looping, and the number of sounds, are all defined here. We are kind of setting the “rules” of how the sound should react to the player’s actions. Sometimes, this can be very easy (think a menu button sound) or very complex (full dynamic mixing system with parameters and runtime rules for reverberation). More often than not, implementation is not just set and forgotten. We usually come back to this step quite often in the production and sometimes we work on individual sounds, sometimes on bigger groups of sounds that should react similarly.

Finally comes the programming phase. This is when we set, in the game, when a sound should play and set the necessary parameters. Things like saying that the menu button should trigger a sound when clicked or telling the audio system the value of the volume slider in the game settings so that all sounds can play at the appropriate volume. Game audio programming is a whole discipline in itself and includes a lot of things like loading audio data in memory, handling CPU usage, and making sure sound spatialization works.

Normally, after all these steps, the sound conceptualized in the beginning should be playing back in the game (after some bug fixing, most probably). An important factor, however, is that these steps don’t necessarily happen sequentially during production. Many of these can be done in parallel with proper planning (again stressing the importance of the concept phase).

Well, Demute is a company that does all of this, after the importance of the concept phase they do the waiting (not a lazy waiting as you could read above), the production of sounds using different media, the implementation of those sounds with the animations and after that something that I thought was not the responsibility of a sound department, the programming phase, that seems to me one of the most important parts, otherwise in Mario we would have the “uh-uh” sound not when we press the button to jump but maybe when we would fall making it an “uh-uh” sad, oh so very sad!

What next? Maybe, the footsteps of characters, the first thing that comes to mind! Let’s go!

Footsteps

According to Pablo, the approach of making sounds for this part of video games is similar to other sounds, always commencing with the context and goal phase and then pursuing said goals, because different approaches to creating these types of sounds come with different goals. For example, “Are we playing a city builder with many different characters on screens walking around? Then we will design the footsteps to sound discreet (not only in volume but in frequency and dynamics too) and make sure they work well when played in high quantity (so enough variation to avoid jarring repetitions). Are we playing a third-person stealth game where the footsteps of our character give information to the player as to how much noise they are making? Then we will concentrate on making the footsteps sound full and precise to fit their goal.

Technicalities

To close up this part of the interview I tried to understand the technicalities of sound creation and sound mixing. According to Pablo, The most important tool we use daily is our DAW (digital audio workstation). It’s the software that lets us construct a sound by combining, editing, and processing different audio sources. If we were to do a sculpting analogy, the DAW is the tools and the audio sources are the clay, stone, or any other sort of material we sculpt with those tools. The DAW we use here at Demute is Reaper.

To gather audio sources to work with (our sculpting materials) there are many different techniques. We will, for example, record sounds in a controlled environment in our studio (usually referred to as “foley”). We can also go outside with a portable recorder and record sounds from our environment (cities, nature, groups of people, cars, etc). More often than not, we will actually buy sound libraries created by more experienced recordists than ourselves (for example, US-based recordists usually have way more experience recording guns than Europeans). We can also create synthetic audio material to work with. These are computer-generated sounds that are created from basic electric shapes and signals. This can get very complex and usually demands a lot of work to get exactly what we want, but is essential to create sound material that is “extra-natural”. Have you ever had the chance to record a dragon or a spaceship? Me neither, so we rely on artificial sounds to help fake those sounds”.

Ambience and Environmental Sound

At this point in my conversation with Pablo, I was curious about how sound immerses players in the gameplay, how sound can dictate a player’s movement and actions, and how sound can construct an environment. My first question was directed at immersion, how they do it with sound and not overwhelm the player so much they feel lost. Pablo says that for the immersion to be correctly implemented, one must consider two parts design and mixing.

“On the design part of things, we have to think about what is the role of the ambient sound and how to build something that fits the uncontrollable nature of video games. The best example is something like a radio on a level. It will be very tempting to decide we want to play some sort of radio broadcast sound, because that’s what radios do right? However, how are we going to handle the possibility of the player just standing next to the radio for hours? If we loop the sound, the player is going to notice that something isn’t right, but we also cannot create hours of radio content. This extends to any kind of environmental sound. We need to think of what really needs sound and how to shape that sound to be as transparent as possible while still giving the information it needs to give.

On the mixing front, we will mostly have to think about when an ambient sound is important. It’s the same in movies. If you focus on a horror movie, you will notice that any time a jump scare is about to happen, the ambient sound slowly fades out to create an oppressive silence. Ambient sounds are very important to give an atmosphere, but also need to go away when relevant. Therefore, if we want to create working immersion, we have to spend time setting up rules about how ambient sounds will work and react to the player’s actions.”

Sound and mixing sounded perfect but one question popped into my mind. Could it happen that even with meticulous design and mixing of the ambient sounds, a player could be misled by them?

“The particularity of audio is that it is always present and can hardly be filtered out. This means that it can act as a powerful guide to lost players. A well-known example of this is present in multiplayer games, where the sounds a player makes will inform the other players of their location and distance. However, “with great power comes great responsibility”. Players will rely on the fact that the information they are hearing is correct. What could be worse than hearing an enemy shooting at you and then realizing there’s no enemy or danger?

An important fact to remember when thinking about this is that players did not work on the game like we did. Meaning that a lot of our assumptions are going to be invalidated when a new player picks up our game. We should keep an open mind as to what guides and what misleads players and make sure to take feedback from them because nobody plays the game the same way and nobody will play it like we did.”

Remember players, audio is powerful!!!

Voice Acting

Now, let’s dive into the topic of the voice acting sub-topic.

Well… who doesn’t remember exactly the voice of their favorite video game characters? This happens because two things merge: amazing gameplay and narrative and great voice-acting and audio mixing. So, let’s try to understand better the process of voice-acting for video games.

When working with voice actors, what are the key considerations to ensure their performances align with the game’s overall sound design?

I’d say there are two things to take into consideration when thinking about aligning with the overall soundscape of the game. First, is technical consistency. Having voices recorded in different sounding environments will make it way harder to make the player believe they all come from the same virtual space. The second one is context and projection. If an actor’s voice lines have to fit into a scene with a lot of shooting and explosions, we need them to be shouting at the recording stage. If we don’t, then these voice lines are going to be very hard to integrate into the overall mix.

How do you handle voice direction to maintain consistency in tone and quality across different characters and scenarios?

We ensure that there’s always a voice director at every recording session and that the voice director knows not only about the technical specifications but also about the game. Then it’s a matter of using our ears and taking the time to make sure everything is ready before we start recording.

What challenges do you face in integrating recorded dialogue with other in-game sounds to ensure clarity and immersion?

The most common challenge is making dialogue work in a lot of different contexts. Sometimes the same dialogue line needs to work in a quiet environment and a very noisy environment. Ideally, we’d try to have two different recordings for such a case, but it’s not always possible, so we have to find ways to make it work through dynamic mixing.”

Music Score

The last aspect that we will cover about sound design is music scores! The music scores! Oh, the memories… the score when Artur Morgan dies, or the one from Tetris, so far apart in time of production but so incredibly accurate on the gameplay that they were attached to.

A music score can transform a video game, iconize a story, can be something that will endure more than the game itself.

So, I asked Pablo a few questions about music scores and how to implement this part into a sound design job for a video game:

How do you collaborate with composers to ensure the score complements the sound design and enhances the player’s emotional experience?

To get the best results, we need a close collaboration during all of production for music. It is a very powerful tool and thus requires special attention. Usually, we will have long meetings with the game director and the composer, or musical director, to make sure that we are aligned exactly on what the music should provide and its place in the narrative/gameplay of every situation. We write lots of documentation and iterate a lot, until we have something that creates a satisfying experience.

Could you explain the process of dynamic music implementation and how it reacts to the player’s actions within the game?

Dynamic music implementation is this idea that the music should follow the player’s actions without sounding like a very bad dj randomly changing tracks every minute. How to make this work greatly depends on the game, but it mostly relies on decomposing the music in different elements (be it different instruments, different parts, different intensities) and turning them on and off “musically”. By “musically”, I mean in a way that makes musical sense : respecting tempo, harmony and generally making sense. Obviously, this can be easier or harder depending on the music style and the game genre. Free jazz and ambient music are way more flexible in the kinds of transitions that work compared to rock or pop. The ultimate goal is to make the player feel like the music has been specially composed for their unique playthrough.

To finish this amazing exchange of ideas with Pablo, there was only one thing left to ask, about the future of sound design and future projects for him and Demure:

Future of Audio Design for Video Games

What were some of the best projects that you’ve worked on with Demute and ones that you can’t wait to work on? What are some emerging trends in-game sound design, and how do you see the future of sound in video games evolving?

We have worked on more than 60 projects by now, so claiming any of those to be the best would be hurtful to the others! From small indie projects like Please Touch The Artwork to huge games like Stellaris, we take a lot of pride in everything we do. We have a couple of games lined up in production that we can’t wait for people to hear and give us feedback on, but I can’t talk about them yet, unfortunately.

The latest trends have mostly focused on spatial audio (bringing more layers of immersion to players with great sound systems at home or in VR) and user-generated content. Obviously, AI is in everyone’s mind right now and it is indeed slowly being integrated in the tools we use and the way we work. We are already using it for dialogue editing and programming for example. I personally can’t wait to see how we can continue working towards less menial work in audio allowing for more space to make creative choices and direction decisions.

The Incredible World of Sound Design For Video Games was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

In this blog, we’ve covered the many contributions that AI has brought to the gaming industry, from generative narrative to our own AI anti-cheat solution, there are many ways that AI has helped shape a more technologically advanced gaming industry.

In this piece, we will examine how AI has helped esports professional athletes and teams become even better at their games.

So, esports athletes train every day to become better at their game of choosing, either getting to know the maps and the movimentations on them very well, or how to use better a weapon or spell, or how to master a character and to use it in different ways to surprise their opponents. These things come from many hours of gaming, study and strategising with teammates and coaches.

AI Tools That Can Help Players Improve Their Performance

With the advent of AI tools for numerous activities, the esports scene also started using these tools to help develop player ability, tactics and other things to improve performance.

Esports video games like CS or League of Legends are, most of the times, decided in the tiniest of details, a pixel to the right or to the left can mean that a game is lost or won, a path choosing or a spell on a certain time might be what differs a team that wins the trophy to the one that keeps on losing in the final. What is all this? This is data, skill too, but data! And what is good at collecting and analysing huge amounts of data? AI.

So, there have been developments of tools that use in-game data to analyze player behavior and upgrade the gameplay of the athletes. The data analyzed by these tools can be of the nature of in-game decisions, match statistics and player movements on the map. With AI collecting and analyzing this kind of data, the team can know where the player missed an opportunity, why the player missed it and most importantly, how the player can improve at each moment of the game.

With esports attracting more and more funding and sponsor deals and with talks that sometime soon esports can become equivalent to sports and even figure in olympic games, constant improvement and game evolution have to go to the next level, if in a “traditional” sport you already do this with football clubs analyzing all the stats collected to buy players, making tactical changes and other actions, esports were expected to do that as well. An example of a tool of this kind is iTero, a League of Legends AI coach that helps players improve their abilities on map movement, champion pick and also builds.

AI Bots

One other topic that can be discussed is AI bots and how they can be used to help esports players. Now, it is good to play against bots, you are no longer considered a noob!! On a more serious note, with an AI capable of producing bots that can learn from real players’ moves and tactics, the world of bots is on a whole different level.

Nowadays, a player can train their skills with these high-end bots because not only do they have human-like behaviors but they can be programmed to play on the highest of levels, meaning that if you can win from these bots, you are one step closer to winning to all your opponents in real life.

AI Serving Esports Broadcast

On a different note, let’s talk about how AI can help the broadcast and content part of esports. The transmission of esports is somehow a new thing, up until now the broadcasts were amateur but they have been more and more professionalized by the official competition holders. The next step is for these transmission rights to start being sold to TV channels and other major broadcasters and with this, the transmission needs to go up in quality. With AI, esports broadcasters can learn how to better trim content for highlights, use past statistics to better comment on the games, and do post-game analysis with a more and more accurate perception of everything that went in the game, even some details that they might have missed during the transmission such as things that were happening on the other side of the map, builds, different paths and many other things.

With AI-tailored transmissions, the possibilities of generating content in each game that is broadcasted are up exponentially as well.

Other Improvements from AI

• Better matchmaking process: with the help of AI, human-made errors can be almost nonexistent on things like keeping records and rankings of players and teams. With AI the matchmaking process will be more tailored and teams and players will get matched with others of very similar abilities and ranking;
• Judging/evaluating players’ abilities: with AI, the process of selecting players to become professional will get more and more precise. With the huge amounts of data and gameplay that AI can analyze, the work of scouting teams on selecting team members will be easier and more accurate;
• Identifying negative player behavior: with AI tools, there will be more monitoring of bad behavior in players while playing the game and in their daily lives;
• Anti-cheat: Anybrain’s AI anti-cheat solution is a good example of how AI is already revolutionizing the gaming industry. This solution can detect cheats with more than 99% accuracy on all games, including the esports ones, assuring a fair and secure competition for all players and teams.

The Game-Changing Role of AI in Competitive Esports was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.

The world of gaming being in constant evolution generates advances in all areas. Every day, the mobile games environment gains traction with more gamers joining this platform. Being the most common and easiest-to-reach platform, it is only natural that games on mobile phones are so popular nowadays. Let’s look at some numbers, for 2024. The mobile games market is projected to generate a revenue of 98.7 billion dollars across the world and the number of users in this segment is expected to reach more than 1.9 billion users by 2027, not only is the market growing exponentially in terms of revenue but it is also growing in terms of gamers. It is expected to only go up from here, transforming it into one of the most populated environments in the world of gaming.

Is there a cheating problem in mobile games?

When something gets more and more popular, it attracts more and more different types of people, from the good to the bad and, in gaming, cheaters are the pinnacle of what you might call bad people in gaming. Cheaters no longer belong exclusively in FPS games and using aimbot, they have found ways to build frauds in mobile games, all types of mobile games, arcade games, snooker games, platforms, etc… Let’s try and see the problem of cheats in mobile games and find solutions to protect these types of video games.

Let’s see what we are up to:

• A few months back we wrote an article about the growing danger of emulators being used to gain unfair advantages in mobile games. These players were using these programs to play mobile games with non-supported controllers or with other types of advantages. You can check our article on the subject to understand this in more detail.
• In this topic, one can also mention modding in video games, a thing that is very common in mobile games. Although it can be understood that modding is not always cheating behavior, when modding is not permitted by the studio that built the game, it is cheating because if you control the mod you will gain an advantage over your adversaries, meaning, cheating. Usually, modding a mobile game, means that you download an IPA/APK files in the form of an app that you will keep running while you run the game and this will grant your account infinite resources or in-game money (depending on which type you are using), it is very simple to use, you don’t require any advanced computer technics, making it one of the preferred way to cheat in this area of gaming.
• But cheating in mobile games is, nowadays, much more than using emulators or modding. There are specific cheaters for mobile games that are used in the same way as one uses an aimbot while playing an FPS game. One simple look at the first page of Google when one searches for “cheats in mobile games” gives you an idea of how this problem is. Look at the image below and then let’s think about this issue.

• Here it is: with one simple search on Google, you can find videos explaining how to hack certain video games, most of the video games for a certain operating system, and you can also visit forums of cheaters where you can find specific cheats for specific games. For example, you can find multiple cheats for Brawl Stars, from aimbots to ESP (Wallhacks), bots to help you farm gems and coins and other variations of these cheats are available simply throughout the internet. A person with advanced knowledge of where to find these cheats or even how to build them will have an even greater advantage in winning every match in a game such as this one. I’ve used Brawl Stars simply as a demonstrative example, I don’t have any data that says that more cheats are being used in this game than any other or that there more cheaters in this game than in any other.

Ramifications of the Problem

So, three ways of cheating in mobile games, hardware cheats (emulators), modding and the “old” cheats that have arrived in mobile games. All three work in different ways and all three can be used (although illegally) in the comfort of your home but the scenario changes when you are playing on a LAN or an official tournament. Hardware cheats will be out of order because they are visible to the human eye as well as modified games but the case might not apply when we think about the other types of cheats (aimbot, ESP…) and this can be a problem. Mobile games are growing and so are their esports! Until some time ago, mobile games esports and tournaments would not get many views and fans but the mobile versions of known games like League of Legends, Call of Duty and others have shifted the tide, mobile games are more and more visible and so are their tournaments. With this, the need for security and anti-fraud in mobile esports grows and is now more needed than ever. But how can one protect mobile games from cheaters? In the same way as one protected a computer or console? Is it easy or difficult? Let’s get into that.

Solutions for the problem

Security for mobile games tries to be nowadays as strong as that of a PC or console game. There is a need for data protection, protection against hackers and also need for anti-cheat.

There are two ways of anti-cheat for mobile games one, we could call it a more traditional way of doing things, almost as one would transfer the traditional way of anti-cheat for PC games to a mobile system, this way focus on checks for integrity on the phone while the game is open but also checks for overlays (to prevent the use of tools that can be used over the game while playing that will grant the player an unfair advantage) and of course, the check of the entire phone for installed applications that can be viewed as cheats or unofficial mods that are not permitted by the game studio. This way of doing things can and does work for some cases but can also be bypassed by players that run the game inside a Subsystem and use the cheat outside of it, putting out of use the anti-cheat since he would have access only to limit information and would not be able to identify anything wrong.

At Anybrain, we’ve been doing things differently. We’ve developed a solution that aims to, not only give players more privacy and control over their data (our solution doesn’t require the full analysis of files on the players’ phone, not even when they are playing the game) but also protect them from encounter cheaters using a different technology. We’ve developed a way of doing anti-cheat that resonates around biometric profiles. With the use of more than 200 metrics (such as the speed of clicks on the screen of the phone, the angle of inclination of the phone while playing the game amongst others), we create a profile that is unique for each player, almost like a gamer fingerprint. Then, when a gamer uses a cheat, their behavior will be different from the data that we’ve collected to form the biometric profile of that player and the player is immediately flagged as a potential cheater. This way of doing things can help protect a game from many frauds, hardware cheats (the emulators that we wrote about at the beginning of this text) but also from all the other cheats, like bots, and the “old” cheats that are now being used on mobile devices like aimbot. Anybrain’s solution works in real-time, from the moment we are present in a game, our machine-learning models work to identify non-human behaviors. This is the ultimate protection against new bot accounts that are created after the ones that are banned. These types of solutions can work side by side with the more traditional one to ensure maximum protection, there is no better solution than the other, they simply work in different ways and they can be complemented.

If you are building a mobile game and want to talk with us about how Anybrain can upgrade the security of your game, sign up here.

Cheats on Mobile Games: new (old) cheats and how to beat them was originally published in Anybrain on Medium, where people are continuing the conversation by highlighting and responding to this story.