Inspiration
Most scams are not sophisticated attacks. They are repetitive, low-effort operations built from reused scripts, fake invoices, impersonation messages, and pressure tactics sent at scale. These schemes continue to succeed because consumers are forced to evaluate suspicious requests alone, in real time.
That should no longer be acceptable. We built the Warden to bring instant, AI-powered scam intelligence to everyday payments and documents, helping users identify fraud signals before money leaves their account.
What it does
The Warden is a proactive security tool that stops scams before money leaves your account. When bunq’s internal risk system, Finn, flags a payment as suspicious, the transfer is moved into a 24-hour safety window instead of being sent immediately. During this 24-hour window, the user can still cancel the transaction before the funds are released.
The Warden connects to the current bunq API flow, receives flagged payments, allows users to upload extra context such as chat screenshots, invoices, or emails, and analyzes that evidence in real time. The system then returns a clear verdict: scam, suspicious, or low risk.
If a scam is detected, the user is guided to cancel the payment while the money is still safe. Rather than trying to recover stolen funds afterward, the Warden prevents fraud at the moment it matters most.
Challenges we ran into
We iterated heavily on prompts to find the right balance between precision and recall. Small prompt changes often had large effects on verdict quality and consistency. Another challenge was making outputs clear and actionable, so users could confidently decide whether to cancel a payment within the 24-hour safety window.
What's next for The Warden
The next step is to improve the scam detection engine with more real-world scam cases, richer transaction context, and stronger multimodal reasoning across screenshots, documents, and conversations. This will increase accuracy, reduce false alerts, and help the Warden detect new scam patterns as they emerge.
Log in or sign up for Devpost to join the conversation.