An AI voice assistant, in development
When something feels wrong, a conversation could make the difference. We’re building ScamCatch to help people recognize and step away from an ongoing scam.
Explore the approachBuilt around a simple idea: you don’t have to figure it out alone.
“Let’s pause for a moment. What did they ask you to do?”
“They said I need to move my money to keep it safe.”
“Before you send anything, let’s check with your bank through a number you know.”
Our goal Give doubt a voice before money is lost.
01 / The approach
We’re exploring a voice assistant that helps someone slow down, explain the situation, and decide what to do next.
Proposed experience
A risk signal, such as an unusual payment flagged by a bank, could prompt a check before the person sends money.
Early signalThe AI would ask the person to pause and explain what happened, then discuss signs of pressure, secrecy, or deception.
A calm conversationThe person could verify the request independently and choose whether to involve their bank or someone they trust.
Support with consentBank signals, bank alerts, and trusted-contact sharing are planned research directions. They are not available features.
02 / Where we are
Can a timely voice conversation help someone pause and verify a request? That is what we want to learn.
Current development prototype
Our Android prototype analyzes on-screen text with the user’s permission. When it flags a possible scam, it can start an AI voice warning.
The voice assistant asks the person to pause before sending money or sensitive information. The user can end the conversation.
Next research direction
We want to explore risk signals from financial institutions, with a voice check before a risky payment.
We also want to explore support from trusted contacts and fraud specialists, with clear choices about what gets shared.
ScamCatch is not publicly available. The prototype does not block payments, and we have not established that it prevents losses.
03 / What matters
We’re designing for people facing pressure, including older adults and the people who care about them.
The conversation should help someone assess the situation without judging them.
A warning can be wrong. The next step should be independent verification, with uncertainty made clear.
People should choose their trusted contacts and understand what information would be shared.
04 / Why we’re building
Watching interviews with scam victims led Jonathan Kim to a question: could an early conversation help someone trust their doubts and pause?
ScamCatch grew from that question. We’re exploring how AI voice could help someone assess a request while there is still time to act.