AI Fraud Decoded 2 | Why Scammers Already Know You Just Picked Up Your Package

You just collected a parcel from a convenience store. Minutes later, your phone rings — someone claiming to be from the logistics company, knowing exactly which platform you ordered from and where you picked it up. It sounds legitimate. That is precisely the point.

This kind of precision is not coincidence. It is the product of a sophisticated underground data economy in which leaked personal information is bought, sold, reassembled, and weaponized — often long after the original breach has been forgotten.

Wang Chih-ching (王志清), co-founder and Chief Digital Officer of digital consulting firm Leadbest Consulting, has spent years tracking cybersecurity threats and dark web activity. His observation is that Taiwanese personal data is widely available on dark web marketplaces, and the nature of what is being traded has changed significantly.

Early data leaks typically involved a single source — a government database, an e-commerce platform, a membership system — being packaged and sold. What is more common now is data from multiple sources being cross-referenced and consolidated. National ID numbers, gender, registered home addresses, mobile numbers, telecom carrier details, e-commerce transaction records, and purchasing histories are gradually layered onto the same individual profile.

Wang describes this not as data stacking, but as assembling a sketch of a person. The more sources that feed into it, the more precisely a fraud syndicate can map someone's daily habits, spending patterns, social connections, and potential vulnerabilities.

Critically, leaked data does not disappear when a breach is announced and users change their passwords. In the underground market, data gets resold, deduplicated, merged with newer datasets, and repackaged. A breach from years ago may still be in active circulation, updated and enriched with information from subsequent leaks. Leaked personal data is not a one-time liability — it is a raw material that gets processed repeatedly.

The dark web data economy is more sophisticated than most people imagine. Wang noted that listings go well beyond basic contact details. Credit card data, for instance, may include the cardholder's name, card number, CVV verification code, expiry date, and a success-rate rating by country or risk category. Hacking tools, system vulnerability exploits, and even targeted bounties — requests to obtain data on a specific individual — are all available.

What this means is that the underground market does not simply sell raw lists. It sells organized, labeled, actionable intelligence. The more complete and current the data, the higher the price and the greater its utility as a fraud instrument.

The market has also evolved from selling lists to selling query services. Just as legitimate services allow users to check whether their passwords have been exposed in known breaches, dark web equivalents allow buyers to query sensitive information on demand. This model means a buyer does not need to process raw data themselves — they simply describe what they need and pay for the result.

Many people assume that serious fraud risk only arises when complete data — national ID numbers, bank account details, credit card credentials — is compromised. But from a fraud execution standpoint, the most dangerous data is often not the most complete. It is the fragment that matches the target's current life situation.

Wang explained that a fraud syndicate with access to e-commerce transaction records does not need a full identity profile. Knowing what someone recently purchased, which platform they used, and how they received the delivery is enough to construct a convincing fake logistics or customer service message.

This is why victims frequently say afterward that they believed the call or message because the caller seemed to know their details. Fraud syndicates exploit what might be called the partial truth effect: when one element of a message is accurate, targets tend to extend trust to the rest of it. A generic alert about a package problem raises suspicion. The same alert, referencing a specific platform and pickup location the recipient actually used, reads like a legitimate service notification.

Fraud does not win through complete fabrication. It wins by mixing real data with a false scenario, leaving no time to tell them apart.

Leaked data has always been dangerous, but AI has significantly lowered the barrier to exploiting it. Previously, turning large volumes of raw data into usable attack material required skills in data mining, programming, and analysis. Identifying relationship networks, spending patterns, or investment preferences from a dataset required a degree of technical expertise that limited who could do it.

Generative AI changes that equation. It can assist with data organization, suggest analytical approaches, write code, and convert raw information into tailored fraud scripts. Wang noted that data mining was already capable of mapping a person's social graph — identifying close contacts and relationships — for use in scripted attacks. AI makes this process faster and more accessible to operators without deep technical backgrounds.

The result is that a leaked phone list is no longer just a phone list. It can be enriched into a set of character profiles: who is likely elderly, who shops online frequently, who follows investment content, who may have a mortgage or car loan, who regularly posts about their daily life on social media. Each profile can be matched to a corresponding fraud script, moving the operation from mass random attempts toward segmented, targeted attacks.

Personal data in the AI era extends well beyond text. Publicly available images, video appearances, and voice recordings have become raw material for deepfake fraud. This explains the proliferation of cases involving fake celebrity investment endorsements, fake doctors selling health products, and fake financial experts recommending schemes — all produced using AI-generated or manipulated media.

Public figures, news anchors, physicians, and financial commentators appear frequently in media, making their visual and audio profiles easy to obtain and use for training generative models. Once a convincing deepfake is produced, it can be deployed at scale to create the appearance of personal endorsement.

More targeted attacks are also possible. If a fraud syndicate has both a person's voice sample and their social network data, it can combine the two to impersonate a family member or close friend in a highly specific scenario. Wang acknowledged that such high-cost attacks are not yet the dominant mode, but the technical threshold is falling and the risk is rising.

When fraud syndicates possess not just phone numbers but purchasing histories, social connections, family dynamics, and contextual life details, standard fraud prevention advice becomes insufficient. Warnings not to click unknown links, not to share one-time passwords, and not to transfer money to strangers remain valid — but they no longer cover the full threat surface.

Fraud messages may now arrive from compromised accounts belonging to actual friends or family members. The link may not come first — syndicates may establish trust through LINE groups, fake customer service interactions, or staged investment communities before any financial request is made.

Cybersecurity firm Trend Micro has noted that digital threats in the AI era have evolved from technical attacks on devices to manipulation of information, judgment, and trust — and that protection should extend beyond the individual to encompass the entire household. This framing reflects an important reality: one person's leaked data can be used to defraud their contacts. A compromised account can be used to target friends. A voice sample can be used to deceive family members. Elderly relatives, children, and young adults newly entering the workforce may have lower familiarity with digital fraud and become the next point of attack.

Practical household-level measures — avoiding password reuse, enabling two-factor authentication on social accounts, ensuring family members recognize screen-sharing requests as high-risk, and verifying urgent money requests from apparent acquaintances — all affect the overall resilience of a household, not just an individual.

The trajectory from leaked data to dark web aggregation to AI-assisted profiling to precision fraud attack is not a series of disconnected events. It is a supply chain, and personal data is the raw material at its origin.

Fraud syndicates do not need to know everything about a target. They need just enough to be credible for one specific moment: the moment someone just picked up a parcel, filed a tax return, attended a concert, joined a new platform, or made a purchase. Place the right fragment of real information inside a false scenario, and the target may not have time to distinguish between the two.

For the individual, a data breach may seem to end with a spam message or an unwanted sales call. For the fraud industry, that same data is a lead, a traffic source, a script foundation, and the opening move of the next precision attack.

Original Article in Chinese

（Related： AI Fraud Decoded 1 | Buying Groceries Online Could Be the First Step to Getting Scammed ｜ Latest ）