The New Face of Banking Fraud: When AI Voices and Fake Pages Join Forces

The landscape of financial fraud has taken a disturbing turn, one that threatens to undermine the very security measures banks have spent years building. Imagine receiving an urgent call from your bank’s fraud department, warning you that someone is trying to access your account. The voice sounds professional, calm, and reassuringly human. You’re told to verify your identity by entering a code sent to your phone, and you’re directed to a website that looks exactly like your bank’s official portal. Every security check feels legitimate because, in a sense, it is—the criminals have found a way to sit invisibly between you and your real bank, manipulating both sides of the conversation in real time. This isn’t science fiction or a theoretical vulnerability. It’s happening right now to customers of more than twenty Mexican financial institutions, and security researchers warn it represents a blueprint for the future of account takeover attacks worldwide. What makes this campaign particularly alarming is that it defeats multi-factor authentication, the very safeguard that banks and security experts have long presented as the gold standard for protecting sensitive accounts. A text message code, it turns out, is only as secure as the person who receives it—and when that person is being actively coached by an AI-powered voice on one end and a convincingly real fake website on the other, the code becomes just another piece of information handed directly to the attackers.

At the center of this threat is a sophisticated criminal platform tracked by cybersecurity firm Group-IB as “Balonx Sistema.” The name may sound benign, but what it represents is a fundamental shift in how banking fraud operates—a shift from amateurish phishing emails to an industrialized, subscription-based crime service. Balonx Sistema operates as a Phishing-as-a-Service platform, meaning that aspiring fraudsters don’t need advanced technical skills or deep programming knowledge to launch devastating attacks. They simply subscribe to the service, choosing from individual or office plans that give them controlled access to victim sessions and a selection of banking brands to impersonate. This business model lowers the barrier to entry dramatically, allowing even low-skilled criminals to run banking scams at scale. Since at least October 2025, the platform has collected credentials and financial data from more than 1,100 victims across Mexico, according to Group-IB’s analysis. The researchers discovered the operation after leaked GitHub repositories exposed critical components of its infrastructure and affiliate network, offering a rare glimpse into the inner workings of a modern cybercrime enterprise. What they found was a carefully engineered system that combines three distinct attack vectors: real-time phishing pages, an Android remote-access trojan, and automated voice fraud powered by artificial intelligence. The result is a blended, multi-stage attack that moves seamlessly from an urgent phone call to a convincing website to, in some cases, a malicious mobile application. It’s a chilling demonstration of how cybercriminals are now weaponizing everyday technologies to create attacks that feel personal, responsive, and utterly convincing to their victims.

The technical sophistication behind Balonx’s live phishing operation is where the platform truly distinguishes itself from older, clumsier fraud schemes. Traditional phishing attacks work by simply collecting passwords and hoping they work. Balonx operates differently—it maintains a persistent WebSocket connection that keeps the fake phishing page and the criminal’s control panel synchronized in real time. This means that when a victim types their banking credentials into what they believe is their bank’s website, the attacker can instantly relay those credentials to the genuine bank site behind the scenes. The real bank, unaware that someone else is pulling the strings, responds by triggering a multi-factor authentication prompt—perhaps sending a one-time code via SMS or push notification. In that precise moment, the Balonx operator sees the request and immediately displays a matching fake verification screen to the victim, asking them to enter the code they just received. To the victim, this all looks like normal banking procedure. To the bank’s security systems, this looks like a legitimate customer authenticating normally. But in reality, the code has been captured and used by the attacker to complete the login. This real-time relay technique, sometimes called a “man-in-the-middle” attack, is far more dangerous than simple credential theft because it defeats the entire purpose of two-factor authentication. The platform even supports fourteen different screen types, allowing the attacker to adapt their deception as the session develops—requesting SMS codes, purchase-approval confirmations, ATM PINs, card details, or cardless-withdrawal codes under the appearance of routine security checks. The operator can change the story mid-flow, guiding the victim through an ever-shifting maze of requests that all seem legitimate in context.

What makes this campaign even more unsettling is the addition of an AI-powered voice component that humanizes the attack in ways traditional phishing never could. Balonx includes a module called CallFlow, which uses a combination of language models, speech-to-text processing, and synthetic speech generation to conduct phone calls as a fabricated bank representative named Carolina. This isn’t a simple pre-recorded message—Carolina is an interactive system capable of holding a conversation, responding to questions, and guiding the victim through the steps of the scam. The platform can run outbound call campaigns without requiring a human operator to be present for every conversation, meaning attacks can scale to thousands of potential victims simultaneously. The voice is designed to sound warm and professional, adding a layer of social engineering that makes the subsequent phishing page even more convincing. When a victim receives a call from “Carolina” warning them about suspicious activity on their account, then follows her instructions to visit a link and enter their credentials, they’re responding to a manufactured crisis that feels personal because the voice on the other end seems to know them. This automation of voice fraud represents a significant escalation in the arms race between banks and criminals. Recent coverage of automated bank support calls has highlighted similar dangers, showing how a synthesized voice can steer a target toward a fake page, extract sensitive information, or even persuade them to install software. In Balonx’s case, the AI phone call and the live phishing page work in concert, creating a seamless patient journey that the victim experiences as genuine customer service but is actually a carefully orchestrated theft.

The attack doesn’t stop at stealing a password, however. Balonx Sistema also incorporates a mobile component that can escalate an account compromise into a full device takeover. Through a fake bank-protection alert screen, the platform distributes an Android remote-access trojan built on the Spyroid framework. If the victim can be persuaded to install this app—perhaps by Carolina’s urgent warnings about needing to secure their phone—the malware establishes a persistent connection to its command-and-control server and begins exfiltrating screen content, keystrokes, SMS messages, and even activity from legitimate banking apps installed on the device. The connection is designed never to time out, meaning the attacker maintains uninterrupted access long after the initial deception succeeds. This transforms the attack from a single-account compromise into a potentially catastrophic invasion of the victim’s digital life. The criminals can observe everything the victim does on their phone, intercept messages, capture passwords as they’re typed, and even hijack banking sessions in progress. This pattern is increasingly common in the Android threat landscape, where criminals use fraudulent websites or phone calls to push apps outside official app stores, bypassing the security scrutiny that Google would normally apply. Bank customers who find themselves in this position are often completely unaware that their device has been compromised until they notice unauthorized transactions or find their accounts emptied. The combination of real-time phishing, AI voice calls, and stealthy mobile malware makes Balonx Sistema one of the most sophisticated and dangerous banking fraud platforms yet discovered.

For ordinary bank customers and financial institutions alike, the emergence of Balonx Sistema demands a fundamental rethink of how we approach account security. The old advice—never share your password, always verify the website URL—is no longer sufficient when attackers can create real-time illusions that perfectly mimic legitimate banking interactions. The most basic line of defense is behavioral: customers should end any unsolicited call claiming to be from their bank, and independently call the number printed on the back of their card or displayed in the official banking app. They should be deeply suspicious of any caller who directs them to a website or asks them to install an app, regardless of how convincing the voice sounds or how official the site appears. Requests for PINs, CVV codes, or card scans should always be treated as warnings signs, because legitimate banks never ask for these through unverified channels. Financial institutions, meanwhile, need to monitor for the infrastructure patterns Group-IB identified, including unusual WebSocket activity, suspicious redirect chains, and the specific domains and IP addresses associated with this campaign. For high-value accounts, banks should consider moving beyond SMS-based authentication to hardware security keys based on FIDO2 standards, which are far more resistant to real-time relay attacks. Anyone who believes they may have fallen victim should contact their bank through verified official channels immediately, reset all credentials, and carefully review recent transactions. The Balonx Sistema campaign is a stark reminder that security is not a static destination but an ongoing process of adaptation—and that the criminals, armed with artificial intelligence and industrialized phishing services, are adapting faster than ever before.