AI Outbound Harassment and Voice Cloning Scams: Mass Dialing Impersonation of Acquaintances or Customer Service to Defraud Money

Who Gets Targeted

Victims are primarily the middle-aged and elderly, as well as individuals with low information sensitivity, particularly elderly people living alone and urban residents lacking awareness of AI technology. Such individuals naturally trust the voices of relatives or authoritative figures like police, prosecutors, or customer service representatives on the phone. Upon hearing a realistic AI-cloned voice, they become anxious, their judgment declines, and they often transfer money or dictate verification codes according to instructions before they can cross-verify. Another group of victims includes ordinary consumers frequently harassed by AI marketing calls; their personal information is bought and sold in bulk through gray market channels. Even if they block numbers, they cannot escape the harassment due to number rotation, leading to psychological fatigue and making them more susceptible to well-designed scam scripts.

骗局怎么运作

• Scam syndicates first obtain citizens' mobile numbers, names, and social relationships in bulk through black market channels. Simultaneously, they use public short videos, live stream recordings, and other materials to extract voiceprint features of target relatives or acquaintances using AI voice cloning models, generating highly realistic voices capable of real-time conversation. This makes it difficult for victims to detect anomalies in tone or intonation, laying a foundation of trust for subsequent social engineering manipulation.
• Some syndicates deploy AI automated outbound calling systems, with a single device capable of making over 1,000 to 1,500 calls per day. They bypass marking and interception through carrier number pool rotation, virtual number masking, and anti-marking gray market services, screening potential victims via carpet-bombing tactics. The system can classify and route calls in real-time based on the recipient's gender, age, and response keywords, automatically transferring high-intent calls to human operators or triggering deepfake script flows.
• During the call, the AI customer service or AI relative applies pressure using emergency scenarios: impersonating e-commerce customer service claiming an order issue requires a refund, impersonating police or prosecutors claiming an account is involved in money laundering, or impersonating a child claiming an urgent need for money due to an accident. The scripts are iterated through multiple rounds, automatically adjusting responses based on the victim's follow-up questions, and even inserting background sound effects like police sirens or hospital noises to enhance urgency, preventing the victim from verifying with family members while in a state of high tension.
• For precision voice-cloning scams, syndicates often first make a silent call to detect the recipient's voiceprint response, then train a cloning model using the collected voiceprint segments before calling back as a relative. The AI-synthesized voice can mimic dialects, coughing, sighing, and other habitual vocal patterns. When the victim hears a familiar voice saying 'Mom, I'm in trouble,' they become extremely panicked and quickly transfer money to the designated account or dictate SMS verification codes as requested.
• Once funds are received, the syndicate quickly splits and transfers them through multi-level bank accounts or fourth-party payment platforms. In some cases, specialized individuals, such as 'runners' impersonating others, collect cash or transfer illicit funds offline. This forms a closed-loop chain from AI outbound calling to voice cloning, remote manipulation, and offline money laundering, significantly increasing the difficulty for police to recover assets. By the time victims realize they have been scammed, the funds have often been transferred overseas or cashed out in small, untraceable amounts.

红旗信号（看到这些快跑）

• 🚩 Incoming calls from unknown numbers, overseas numbers starting with '00' or '+', or numbers frequently changed despite being marked as harassment or fraud by others. A 1-2 second silent delay after answering is a typical technical characteristic of AI voiceprint collection or transfer to a human operator.
• 🚩 The conversation rhythm is abnormally mechanical, with vague or repetitive responses to open-ended questions, stalling tactics like 'What did you say? I can't hear you,' or unnatural shifts in background noise, suggesting technical bottlenecks in AI-synthesized voice processing during multi-turn interactions.
• 🚩 The caller claims an emergency such as an order issue or legal case, but the required procedures clearly violate standard institutional protocols, such as requesting screen sharing, downloading unknown apps, transferring money to personal accounts, or dictating SMS verification codes—all typical signs of a scam.
• 🚩 The caller's voice sounds like a relative but only mentions vague identity information (e.g., 'It's your son') without volunteering full names or family details. They avoid or hang up when asked specific questions, or repeatedly urge you to 'transfer money quickly, don't hang up, don't tell anyone,' deliberately creating information isolation.
• 🚩 Your phone continuously receives marketing or refund calls from different numbers belonging to the same entity, with a frequency of over 5 calls per day, or the same script appears repeatedly across different numbers. This indicates your number has been added to an AI outbound calling pool; you should immediately enable carrier high-frequency interception and anti-fraud app call alerts.

真实案例

• In July 2026, Fuzhou Minhou Police received a report of an AI voice-impersonation scam. An elderly person received a call from their 'grandson' claiming to be in trouble and needing money. The voice was highly realistic, and the family nearly transferred the money before police intervened to stop the scam. This case is a typical example of using AI voice cloning to impersonate family members to defraud the elderly, reported by Fuzhou News on July 1, 2026.
• In February 2026, media reported an AI voice-impersonation scam where an individual used AI to clone a victim's relative's voice to conduct phone fraud and impersonated someone to collect the illicit funds in person, eventually being sentenced by a court. The case revealed a complete criminal chain of remote AI voice synthesis combined with offline money collection, reported by Tencent News on February 27, 2026.
• According to investigations by CCTV News and Xinhua Net, in 2026, multiple regions saw cases of scam syndicates using AI customer service systems to make thousands of calls daily, with single syndicates making up to 1,500 calls per day. They used number masking and anti-marking services to evade interception, defrauding victims under the guise of e-commerce platform refund claims. This investigation was reprinted by Sina News from CCTV reports in 2026.
• In April 2023, in the AI face-swapping and voice-impersonation scam involving Mr. Guo from Fuzhou, reported by the Baotou Public Security Bureau in Inner Mongolia, the victim transferred 4.3 million yuan in two transactions within 10 minutes after verifying the identity of a 'friend' via WeChat video call. 3.3684 million yuan was later recovered, with the remaining 931,600 yuan still being pursued. (Source: https://www.zaobao.com.sg/realtime/china/story20230522-1397149)
• In February 2024, Hong Kong police disclosed an AI multi-person face-swapping scam targeting the Hong Kong branch of a multinational company. A victimized employee mistakenly trusted instructions from a forged video conference of executives, transferring a total of 200 million HKD to 5 local bank accounts in 15 transactions. Everyone in the meeting except the victim was an AI forgery, marking Hong Kong's first AI multi-person face-swapping scam. (Source: https://m.mp.oeeee.com/a/BAAFRD000020240301916066.html)

Official Stance

• In 2026, police in multiple regions issued emergency alerts, warning that silent unknown calls hide AI voiceprint collection traps. Criminals first call to collect the victim's voiceprint before conducting cloning scams, urging the public to remain vigilant against unknown calls and avoid answering or staying on the line for long. This alert was reprinted by Toutiao in 2026.
• In April 2026, Xinhua News Agency published a special investigative report titled 'Beware of AI Customer Service Phone Scams,' revealing how the AI outbound calling industry operates to impersonate customer service for refund scams, reminding consumers to verify through official channels and not trust caller identities lightly. The report was published by Xinhua Net on April 28, 2026.
• In July 2026, Yunnan Net published an investigative report titled 'The Proliferation of Thousands of Daily AI Outbound Marketing Calls,' focusing on the status of AI outbound systems being widely abused as tools for harassment and fraud, calling for strengthened number marking and carrier interception mechanisms. The report was published by Yunnan Net on July 29, 2026.
• In August 2026, CNR News reported that over 90% of scams are hidden in apps and other online tools, with AI outbound calling and deepfake technology becoming core means of new types of fraud. The Ministry of Public Security reported over 170 cases involving AI tools in the first half of the year. The report was published by CNR News on August 5, 2026.

How to Protect Yourself

• ✅ Activate high-frequency harassment call interception services provided by carriers (such as China Mobile's high-frequency anti-harassment, China Telecom's Tianyi anti-harassment, or China Unicom's phone manager), and enable unknown call identification and marking features at the phone system level. Directly reject silent calls and overseas numbers to leave no time window for AI voiceprint collection.
• ✅ Install and enable the call alert feature of the National Anti-Fraud Center App to ensure real-time interception of marked scam numbers. Regularly update the app to obtain the latest scam number database and identification rules, and enable dual-channel protection through call alerts and app self-checks.
• ✅ Agree on a safety code or exclusive verification question for calls with family members. For any relative's call involving money transfers, always hang up and call back using a pre-saved number to verify. For calls from police, prosecutors, or customer service, always hang up and cross-verify through official apps, websites, or customer service hotlines; never perform operations directly during the call.
• ✅ Reduce the public posting of videos featuring yourself or family members speaking on social media and short video platforms to prevent voiceprint materials from being scraped for AI cloning training. Regularly check if your personal mobile number has been leaked through delivery or food ordering services, and replace the number or use a virtual secondary number for registration scenarios if necessary.