Antikk is a Cognitive BreakEngine designed for the Touch 'n Go (TNG) eWallet. It is a three-layer system designed specifically for the Silver Economy and high-risk cohorts that doesn't just detect fraud risk—it actively forces a pause, asks "why," and brings a trusted guardian into the loop to prevent scams in real time.
Malaysians Lost RM2.8 Billion to Scams in 2025. Is BNM's Response Matching the Crisis?
Malaysian police are investigating over 450 deepfake scams involving voice impersonation, resulting in millions of Ringgit in losses. Scammers use AI to mimic the voices of victims' acquaintances, convincing them to transfer money to mule accounts. In early 2026 alone, an Assistant Manager lost over RM1.3 Million to a highly coordinated phone scam.
How does scam accountability look in practice, and more importantly, are enough measures being put in place to protect Malaysians and their money?
Current preventative measures rely on reactive or manual steps:
- NSRC 997 Hotline: A 24/7 hotline to freeze accounts and trace funds after the money is gone.
- Semak Mule (PDRM): A portal to check if a recipient account or phone number is a known scammer before transferring, relying entirely on user diligence.
95% of fraud cases are Authorized Push Payment (APP) fraud.
Scammers don't hack the app; they hack the human. They convince the user to press "Confirm" while in a state of PANIC. The current systems fail because they expect a panicked victim to make a rational risk assessment in the heat of the moment.
Antikk is a three-layer system that doesn't just silently detect risk behind the scenes. It intercepts the user experience, de-escalates panic, and forces a cognitive break.
Score the transaction. Talk to the user. Loop in a guardian.
- < 50ms latency per-transaction.
- Assesses recipient trust and contact-list age.
- Uses a weighted ensemble formula:
R = W1·Tr + W2·Bl + W3·Gc + W4·Bt
- Multilingual / Manglish support with Semantic Scoring.
- Semantic Voting: AI asks identifying questions ("Who is this? Are they pressuring you?").
- Emotion Detection: Prosody-based detection to identify panic in the user's voice.
- Cognitive Break: De-escalates the situation and actively forces the user to pause and think during a high-stress scam.
- Protects the Silver Economy & high-risk cohorts.
- Places suspicious transactions under a "Report & Hold" model.
- Requires pre-registered guardian approval before the transaction is released.
- Provides transparent reasoning to both the user and the guardian.
- Fully compliant with PDPA 2024 amendments.
While Antikk is designed to be built natively into the TNG eWallet, our current prototype is hosted on AWS Amplify.
- Frontend Hosting: AWS Amplify
- Backend Processing: AWS Lambda
- Database: Amazon DynamoDB
- LLM Engine: Qwen (Alibaba)
- Voice / Audio: ElevenLabs (STT & TTS)
- Telephony: Twilio
- User Transfer: The user initiates a transfer via the TNG interface.
- First-Tier Trigger: The frontend triggers AWS Lambda, which fetches risk definitions from Amazon DynamoDB and computes the Layer 01 Risk Score.
- AI Analysis & Scripting: If the transaction is flagged as high-risk, the context is sent to Qwen, which provides deep details analysis, explanation of the risk, and generates a dynamic conversation script.
- Voice Agent Activation:
- ElevenLabs processes the text into speech (TTS) and decodes user replies (STT).
- Twilio immediately calls the user via phone to intercept the transaction.
- Real-Time Interception: As the user speaks to the voice agent, the transcript is continuously evaluated. The AI assesses semantic logic and emotional panic.
- Guardian Intervention: If the AI confirms a high scam probability, the transaction is formally paused. Twilio fires an SMS/notification to the pre-registered guardian (e.g., their children) to review and block the action.