Research

AI Scam Intelligence Platform

One of the laboratory’s first flagship projects. Fragmented leads from news, social media, the dark web, blockchain and mobile channels are organised into a queryable, assessable scam-intelligence base that underpins the daily briefings and deep reports.

Capabilities

• Scam knowledge graph and ontology: scam types, scripts, accounts and money trails modelled as a relationship network
• Multi-channel collection: news, social media, dark web, blockchain and mobile
• Case-level analysis: reconstructing the flow and conversational stages of individual scams
• Daily briefings: Hong Kong, Asia and Global editions produced every day
• Risk assessment and data-sharing framework: graded risk labels with clear sharing boundaries

LLM-Powered Anti-Scam Interaction

Using large language models for conversational intervention: stopping victims mid-scam, engaging scammers proactively to gather intelligence, and providing follow-up support for those already defrauded.

Capabilities

• LLM-driven anti-scam dialogue: explaining risk and persuading users to stop
• Undercover AI dialogue: engaging scam operations to collect evidence and extract scripts
• Risk alerts and outbound outreach to high-risk individuals
• Counselling and correction: intervention design for victims and deeply conditioned targets

Police Social Media Data and Communication Strategy

Providing social-media analytics and communication strategy to the police, and moving research from after-the-fact warnings to field experiments on pre-exposure “inoculation”.

Capabilities

• Analytics of the Force’s social media account performance
• Communication strategy and content design
• Narrative-inoculation field experiments: pre-exposure to weakened scam tactics
• Structuring and profiling online-shopping scam samples

Public Opinion Report on Telecom and Online Financial Scams in Hong Kong

Using Hong Kong’s four financial regulators (HKMA, SFC, MPFA and IA) as an analytical frame, the report examines financial scams reported in the news and on social media over the previous year, identifying trends and manipulation mechanisms and setting out policy recommendations and risk warnings.

• Four dominant typologies: counterfeit crypto, impersonation of banks or financial institutions, fake securities and fund sales, and payment induced on fraudulent trading platforms — counterfeit crypto being the most common
• Scams unfold in three stages — initial contact, trust building and action inducement — with scripts and psychological pressure escalating at each stage
• Telegram’s anonymity and community features make it the relay for luring, private coordination and money movement
• Segment-specific manipulation: older victims trust authority, middle-aged victims trust returns, younger victims trust emotion; men encounter counterfeit crypto, women encounter impersonated institutions
• Mainland-born residents in Hong Kong are a high-risk group, being less familiar with local law and more sensitive to official identities
• Scripts are highly formulaic: “zero-risk investment”, “account unfreezing fee”, “celebrity-endorsed recommendation” and “paid-task” phrasing recur constantly

Report available on request

2025 Summer Xiaohongshu Anti-Scam Campaign — Summary Report

A record of a deliberately de-officialised public-interest campaign: 820 creators contacted, 203 publishing, with anti-scam reminders embedded in the rental, banking and pre-departure guides that incoming students actually read — validated against both creator back-end data and a student survey.

• 203 creators published, with estimated total impressions of 2.66 million, 440,000 views and over 8,000 shares
• From 144 back-end datasets collected: about 13,140 impressions, 2,202 views and 42 shares per post
• Survey: 78.5% of incoming students had seen anti-scam content on the platform, and 64.7% noticed the #我在香港防詐騙 tag
• More than ten observed cases of creators interrupting an active scam in the comments
• Execution challenges: platform keyword moderation, machine-only appeals and organised malicious reporting by scam operators

Report available on request