For this interview, we spoke with Ana María Sancho, Chief Artificial Intelligence Officer at Kriptos. Ana has loved math since childhood, which led her to study a degree in Economics—but along the way, she discovered her true passion lay elsewhere. She later pursued a specialization in Big Data and Business Analytics at IE Business School in Spain.
We discussed the importance of AI in preventing human errors in companies—specifically, how it helps reduce information leaks caused by employees or internal collaborators.
Understanding Human Error in Information Leaks
Before diving in, it’s important to identify the types of human errors that can lead to data breaches and harm a company’s operations, reputation, and image.
There are two main types:
- Intentional errors
- Unintentional errors
Want to learn more? Check out our blog: "Employees: Are Yours Secure?"
How Can AI Help Detect and Prevent These Errors?
According to Sancho, certain AI algorithms are especially useful in strengthening administrative controls by recognizing patterns. Identifying unusual behaviors can provide valuable insights that help predict potential data leaks.
Here are some of the key highlights from our conversation:
Are There Different Types of Artificial Intelligence?
Yes. The most common is Narrow AI, which performs specific tasks based on the inferences of a single model.
There are also two more advanced types:
- Broad AI: A system that performs various tasks using multiple models (e.g., self-driving cars).
- General AI: An intelligence capable not only of learning but also of reasoning. This currently exists only in theory.
So Ana, How Can Artificial Intelligence Enhance Cybersecurity Tools?
By detecting anomalies—in other words, identifying irregular user or document behavior and triggering alerts about potential leaks. AI can also identify patterns that indicate which documents are confidential.
Using algorithms, it’s possible to pinpoint the exact characteristics that make a document sensitive—with remarkable precision.
Ana explains that Machine Learning (ML) algorithms are the best way to uncover patterns, optimizing both time and cost while delivering high accuracy.
What Are Supervised Machine Learning Algorithms?
These are systems that simulate human learning to understand the cause-and-effect relationship behind an event.
Take, for example, someone trying to learn what characteristics define a confidential document. They would have to read many known confidential documents to identify patterns and insights—allowing them to recognize new confidential documents in the future.
When a computer performs this same process, it’s called Machine Learning: iterative systems analyze data until they uncover patterns, which are then transformed into logical, mathematical formulas that generate predictions (inferences).
Because ML can analyze more data in less time—and detect every possible pattern—it’s more accurate than human analysis.
Common ML algorithms include:
- Decision Trees
- Linear Regression
- Logistic Regression
- Support Vector Machines
- Perceptrons
- Neural Networks
How Do These Patterns Become Actionable?
Once trained, an algorithm produces a model—a set of parameters that determines the confidentiality level of a document.
When the model is fed new data, it makes an inference, predicting the document’s classification (e.g., confidential, public, internal).
Integrating a precise and well-trained model into a software system enables that system to make decisions and take actions automatically. This is one of AI’s most powerful roles—minimizing human error, saving time, and reducing costs.
How Is the Accuracy of These Algorithms Measured?
Accuracy depends on the quality and relevance of the data used to train the model. For example, if your company is in the banking sector, but the model was trained using data from another industry, its predictions will be less accurate.
Machine learning is all about recognizing patterns, so training data needs to reflect the specific environment you’re working in.
One way to improve precision is by feeding the model a custom dictionary of related terms. In the banking example, if the model was trained in a different sector, the dictionary helps it learn banking-specific terminology and context.
Final Thoughts from Ana
“I truly believe AI is a key tool for preventing human errors. It’s something IT managers can use to anticipate risks and implement proactive controls.”
Ana adds that with the increasing rate of cyberattacks, it’s more important than ever to view AI as an ally.
“It’s a powerful defense mechanism, an anomaly detector, and a smart way to locate confidential information before it becomes a liability.”
