AI Fraud Decoded 7 | Inside Taiwan's 165 Anti-Fraud Hotline, Where Human Judgment Still Outpaces AI

Taiwan's anti-fraud task force was formally established in July 2022. Since then, the Executive Yuan has rolled out two updated versions of its national anti-fraud action plan, and the Legislative Yuan passed a sweeping package of four anti-fraud laws in July 2024. By late August 2024, the National Police Agency's public "165 Anti-Fraud Dashboard" went live, publishing daily updates on the five most common scam tactics and the scripts syndicates use. Around the same time, the agency overhauled how it counts fraud cases and upgraded its reporting systems. This change caused recorded fraud numbers to spike overnight, but also brought official statistics closer to what the public had long suspected was the real scale of the problem.

More than three years after the task force's founding, its dedicated Fraud Crime Prevention Center, operating under the Criminal Investigation Burea, was formally codified into law in late November 2025. What began as an ad hoc task force of just 15 members has since grown to a staff of 136. Yet in a striking contrast, just weeks before the task force's fourth anniversary — at a moment when fraud syndicates have already industrialized their use of artificial intelligence — the Ministry of the Interior quietly announced that the National Police Agency is only now planning to roll out AI tools to assist officers, in phases, starting in 2027 and 2028.

Taiwan's fight against fraud actually predates the current task force by two decades. The 165 anti-fraud consultation hotline launched in 2004, making it the world's first dedicated telephone hotline for fraud reporting at the time — a genuinely pioneering effort. Today, landlines have all but disappeared and smartphones are ubiquitous. Yet, the 165 hotline has survived multiple equipment overhauls and system upgrades while its core operating logic has barely changed in 22 years. According to sources familiar with the system, it remains a point of pride for officers — and reportedly a source of genuine surprise for visiting American police officials touring the facility.

The 165 hotline's headquarters sits tucked into a small hillside building belonging to the National Police Agency's First Security Police Corps in Beitou District, Taipei. On a recent afternoon around 3 pm, aside from a handful of empty seats from staff who had just finished a shift, the office hummed with the sound of 24 simultaneous phone lines — officers confirming details, calming distressed callers, and typing rapidly into their terminals as new information came in. In the summer heat, officers wear specially designed polo shirts rather than standard uniforms, balancing comfort with a sense of order.

Chang Yueh-han (張躍瀚), secretary of the Criminal Investigation Bureau's Fraud Crime Prevention Center, told Storm Media that the 165 hotline currently runs on 53 officers rotating across three daily shifts, operating 24 hours a day, 365 days a year. The hotline fields an estimated 1,500 calls daily, with each officer handling 40 to 50 calls on average. Shift scheduling is adjusted dynamically based on which scam tactics are currently trending and the peak and off-peak hours when the public is most likely to call in.

Since July 2025, fake online shopping scams have become the single most common fraud type reported nationwide, with syndicates timing their scripts to current events and holidays. Ahead of the Dragon Boat Festival, for instance, fake advertisements for rice dumplings appeared online, luring shoppers to place orders before being told their "identity verification was incomplete," their "order settings were incorrect," or "a payment anomaly" had occurred — prompting them to contact customer service, which then walked them through online banking transfers. Because fake shopping scams tend to peak between 8 am and 4 pm, the hotline concentrates more staff during those hours.

An officer surnamed Huang (黃), who answers calls on the 165 line, told Storm Media that the first step is determining what stage of a scam the caller is in. If someone calls before being defrauded, officers provide guidance or help file a report, working to identify the fraud tools involved — phone numbers, email addresses, fake URLs, or mule accounts — before passing the case to the relevant unit. If a caller has already lost money, officers take an initial report, documenting the caller's account in detail before referring them to their nearest local police station to formally file the case.

Despite having a well-rehearsed procedure, Officer Huang noted that many callers arrive in a heightened emotional state — anything from panic and fear to outright anger — and officers must read between the lines of how someone is asking questions to determine what information they actually need. On days with over 100 calls, officers also field a fair share of calls that turn out to be ordinary telemarketing complaints, legitimate loan-brokerage contracts mistaken for scams, or even personal civil disputes that have nothing to do with fraud at all.

Beyond the occasionally absurd calls, Officer Huang said the genuinely hard part of the job is building enough trust in a single phone call for a caller to believe — and accept — that they are actually being scammed. Taiwan's highest-value single-case losses currently come from fake investment scams, where syndicates typically spend weeks or months cultivating a relationship with their target: chatting in social media groups or private messages, sharing fabricated profit screenshots and success stories, and even arranging small withdrawals to build credibility before going for the larger payout.

Undoing months of manufactured trust in a few minutes on the phone is, unsurprisingly, extraordinarily difficult. But that is far from the only challenge facing hotline officers. Huang explained that when a victim reports having already wired money, the priority becomes immediately notifying financial institutions to freeze the transaction for a 48-hour holding period, preventing cash couriers from withdrawing the funds. The problem is that distressed callers sometimes cannot clearly describe the transaction details needed to file that freeze request in time.

Separately, fake police and prosecutor scams — a tactic that has recently resurfaced — typically fabricate a case out of thin air, present forged legal documents, and pressure targets into a panicked confession. Some variants even open with fake "anti-fraud awareness" messaging to build false credibility, before invoking investigative confidentiality rules to isolate the victim from seeking outside help, ultimately extracting credit card numbers, online banking credentials, or securities account information. In these cases, Huang said officers will sometimes ask callers to visit a police station in person rather than rely on a phone explanation, since face-to-face guidance makes it easier for people to recognize what is actually happening to them.

After two years on the hotline and tens of thousands of calls answered, Officer Huang observed that virtually every fraud type follows the same underlying structure, simply dressed up in a different "script." Fake government agency emails resurface as tax-payment phishing texts during filing season; fake lottery-win emails appear around invoice draw dates at month's end. "It's the same method, the same approach," Huang said. "They just package it under a different name — but underneath, it's all the same thing."

Lin Shu-li (林書立), a former chief of the Criminal Investigation Bureau's Crime Prevention Division and now an assistant professor in the Department of Crime Prevention and Corrections at Ming Chuan University (銘傳大學), analyzed the broader trend: fraud syndicates are industrializing alongside technological progress, applying not just AI-driven deepfake techniques but psychological tactics rooted in social engineering. Fake police and prosecutor calls may now be enhanced with increasingly sophisticated AI deepfake audio. Even the caller ID displayed on a 165 hotline call is not immune — syndicates can spoof caller ID numbers using telecom fraud techniques, which is why officials recommend that anyone uncertain about a call's legitimacy hang up and independently redial 165 to verify directly with police.

Here lies the paradox: while fraud syndicates have already reached a high level of sophistication in their AI use, the National Police Agency only presented its roadmap — titled "Tech-Enabled Joint Defense Against Fraud: A New AI Vision for Anti-Fraud" — at a Ministry of the Interior departmental meeting on May 28, outlining plans to introduce AI tools for the 165 hotline in phases between 2027 and 2028. The planned capabilities include real-time call transcription, automated case summary generation, real-time lookup support and response suggestions for officers, and eventually AI-powered voice customer service and automated form-filling — all aimed at reducing the manual documentation burden on hotline staff.

Police have made targeted progress on other fronts in the meantime. To combat fake online shopping scams, authorities coordinated across agencies to halt DNS resolution for fraudulent URLs and pushed financial institutions to shorten the validity window on payment codes. To fight fake police and prosecutor scams, they launched Project Clean Voice (淨話專案), suspending fraudulent phone numbers, seizing related equipment, and shutting down operational sites. To disrupt cryptocurrency-based money laundering, police worked with the Financial Supervisory Commission (金管會) and cryptocurrency exchanges to ban cash purchases of virtual assets, cutting off a key laundering channel. So why has the rollout of AI tools specifically for the hotline moved so slowly?

Chang explained that because much of the data held by police agencies is classified or involves sensitive personal information, the Criminal Investigation Bureau is currently building dedicated AI computing infrastructure to securely integrate with existing police systems, with plans to develop both an internal AI case-assistant tool and a field AI investigative assistant. He noted that the government is also preparing a third version of its national anti-fraud action plan, which would continue allocating budget and resources — covering everything from AI deployment to data analytics procurement, intelligence-gathering tools, forensic software, and cryptocurrency flow-tracing tools.

Notably, since the Fraud Crime Prevention Center became a permanent fixture in November 2025, its structure has expanded well beyond the 165 hotline team. The center now also includes a Telecommunications and Internet Section, responsible for gathering and analyzing data on telecom fraud and fraudulent advertising; a Financial Section, which serves as the liaison point with banks and financial institutions and handles related intelligence analysis; and a dedicated Fraud Investigation Unit (詐欺偵查隊) responsible for actively working cases — combining internal data analysis with field investigation.

While the bureau's centralized AI computing infrastructure remains under construction, sources indicate that the Telecommunications and Financial sections already use existing AI tools informally to analyze trends in de-identified case data. Police sources noted that due to personal data protection constraints, AI currently functions only as a supplementary tool rather than a fully integrated system — full data integration awaits completion of the police agency's planned system architecture. "When it comes to adopting AI, government agencies will naturally lag a bit behind the private sector," one police source acknowledged, "but we're working hard to catch up."

In practical terms, until that computing infrastructure is fully built out, Taiwan's anti-fraud effort remains heavily dependent on frontline officer experience and judgment. To some extent, what currently sustains the national anti-fraud task force is still old-fashioned human expertise rather than artificial intelligence. Fraud syndicates have already put AI to practical use to boost their efficiency; police AI systems remain a work in progress, and whether AI ultimately becomes a decisive tool in the next phase of anti-fraud work remains to be seen. For now, the last line of defense is still the officer on the other end of the phone line — and the investigators quietly analyzing intelligence and tracing money flows behind the scenes.

​Original Article in Chinese​

（Related： AI Fraud Decoded 6 | Why Taiwan Phone Numbers Are Black Market Gold ｜ Latest ）