Cybersecurity: How to Protect Yourself from Scams and Deepfakes

From apps that collect personal data to fake videos generated by artificial intelligence, Professor Giacinto explains digital risks and strategies for protecting your online identity. There cyber security It also comes from everyday habits. A message that mimics a banking communication, an app downloaded from an unofficial source, or a fake video created with artificial intelligence can pose a threat to personal data. Professor Giacinto spoke to Unica Radio about this, explaining how research identifies anomalous device behavior, analyzes deepfakes, and studies the vulnerabilities of artificial intelligence systems. Cybersecurity: The first risk is haste Smartphones and operating systems integrate advanced protection tools. However, technology isn't enough if the user trusts the wrong person. According to Professor Giacinto, one of the main problems is the speed of communications, often accompanied by urgent requests that push users to act without verifying the sender's identity. A common example involves messages that appear to come from the bank and ask the victim to authorize a transfer or block a suspicious transaction. These are social engineering techniques that exploit fear and urgency to induce the victim to take action. The danger increases when the message encourages you to install an application outside of official channels. A malicious program could collect credentials and personal information, allowing attackers to attempt to access your email or bank accounts. How to recognize abnormal app behavior In the research work on cyber securityAn important element is analyzing the network traffic generated by smartphones and computers. Devices continuously communicate with online services and applications: these activities form a sort of habitual profile, which can change based on each user's needs. Experts look for signs that deviate from this behavior. A phone generating unusually high traffic at night, when it's normally not in use, or a device contacting Internet addresses associated with malicious activity can raise suspicions. The analysis also covers application functionality. Access to the microphone, camera, and other sensors should be consistent with the program's functions. An app that activates the microphone for no apparent reason may be a sign that warrants further investigation. Detection systems, however, are faced with increasingly difficult-to-detect attacks. To avoid detection, some malicious programs attempt to mimic the behavior of legitimate applications and perform their activities slowly, without generating obvious anomalies. Deepfake: Why the human eye can be fooled Artificial intelligence allows us to create increasingly realistic images, audio, and video. Early deepfakes often featured visible imperfections, such as unnatural facial movements or unconvincing lip-synching. More recent techniques make these clues much less obvious. As Professor Giacinto explains, specialized software can detect differences that are difficult to see with the naked eye. However, with the improvement of generative tools, analyzing a single image may not be enough. To verify content, it will therefore become increasingly important to compare it with other sources: authentic videos of the same person, background characteristics, habitual movements, and specific facial expressions. This approach integrates technical analysis with fact-checking. The interview also highlights the case of fake videos attributed to real people, in which claims were made that were never actually made. Such content can even deceive family members and acquaintances, demonstrating the importance of verifying the origin of what you watch and share. Possible future solutions include the introduction of digital markers, similar to watermarks or signatures, capable of signaling to machines that content has been artificially generated. Their effectiveness will also depend on the widespread adoption and reliability of the standards adopted. Adversarial Attacks: When Artificial Intelligence Misinterprets Reality Another threat concerns adversarial attacks, techniques that exploit characteristics learned by algorithms to trick them into misclassifying content. The professor offers the example of systems installed in cars to recognize road signs. An algorithm trained on numerous images can learn to distinguish a stop sign from other signs. However, if it relies primarily on statistical characteristics that don't necessarily correspond to those considered important by a person, it can be fooled by an anomalous image. The result is that content that appears to be different from a stop sign could be classified as such by the system. The same principle can apply to images, text, and audio. These limitations show why simply training a model on large amounts of data is not enough: it is also necessary to test its robustness, study potential manipulation attempts, and design tools capable of resisting deceptive inputs. Artificial intelligence: who has the advantage between attackers and defenders? Artificial intelligence is used both to identify vulnerabilities and to develop defense tools. Researchers can simulate attacks to test system resilience, identify anomalies in network traffic, and improve detection capabilities. The attacker's advantage, Professor Giacinto observes, stems primarily from the fact that it only takes one weak point to compromise a system. Defenders, on the other hand, must seek to identify and protect all possible vulnerabilities. AI can make fraudulent messages more credible, generate deceptive content, and speed up the search for information on potential targets. It can also help identify organizations using outdated software or outdated technologies. On the defense front, the scientific community has been working for decades on the use of artificial intelligence for security. The challenge, therefore, is not only to develop more powerful algorithms, but also to reduce system weaknesses and keep infrastructure up to date. Three habits to protect your digital identity During the interview, Professor Giacinto indicates some concrete precautions that everyone can adopt to improve their own health. cyber security. The first is to avoid rushing. When faced with a message announcing an urgent problem, requesting payment, or inviting you to install an app, it's best to stop and check through official channels, without automatically using the links you receive. The second concerns login credentials. It's advisable to use long, unique passwords generated through a password manager, avoiding reusing the same combination across multiple services. When available, it's helpful to enable additional authentication systems, such as those used by banking services and digital identities. The third is to maintain a critical attitude towards websites, offers, and messages. A free service isn't necessarily dangerous, but overly tempting promises, unusual requests, and pressure to make an immediate decision should prompt you to double-check the source. The basic rule is simple: before clicking, downloading, or authorizing an operation, you need to verify. Digital awareness remains an essential component of protection, even as technological tools become increasingly sophisticated. The quest for a safer digital world The Unica Radio interview highlights a challenge facing users, researchers, and technology companies. Cyber ​​threats are evolving rapidly, but the ability to recognize anomalous behavior, verify information, and adopt best practices can significantly reduce risks. Artificial intelligence is part of this challenge: it can be used to deceive, but also to identify anomalies and strengthen defenses. Therefore, alongside research, it remains essential to develop a conscious use of technologies and not rely solely on the protection offered by devices for security. The full interview is available on the Unica Radio website and on Spotify and Amazon Music.