Attackers used to send a link. Now they send “hey, are we still on for golf?” and wait days, sometimes months, before asking for money. That shift is the subject of this conversation with Ariana Mirian, co-founder of BeeSafe.ai, a company that sends undercover AI agents into the same channels attackers use to catch them before they reach a real victim.
Ariana built her career on measurement, starting with a PhD at UC San Diego studying whether phishing simulation training actually works. It largely doesn’t. That rigor now drives BeeSafe’s approach: engaging scammers directly to trace the full attacker lifecycle. Her pilot data found trust-based scam conversations average eight days before a financial ask and go to great lengths to establish a relationship with victims. Many scams are organized crime operations, some staffed by trafficking victims. As Ariana says, there are often victims on both sides of the scam.
Ariana and host Heidi dig into trust creep, the small asks that quietly earn an AI tool full access to your accounts, and how a multi-layered defense is required to combat fraud. BeeSafe operates near the last layer, stopping the money before it disappears.
As Ariana puts it, scam operations are still human scale for now. What happens when they aren’t?
Topics covered
0:00 - Introduction
1:15 - How email deception evolved into slow, trust-based scam conversations
3:00 - What phishing simulation training actually accomplishes (not much)
6:10 - Why this works: billions in fraud losses and a room where most hands go up
10:10 - Inside BeeSafe.ai’s approach to engaging scammers with AI agents
11:55 - Eight days: what the BeeSafe pilot revealed about scam conversation length
13:15 - Why people hand AI tools information they’d never give a stranger
18:20 - Defining trust creep, from scam texts to everyday AI permissions
20:00 - How incremental asks quietly expand an AI tool’s access
21:30 - Pig butchering scams and the organized crime operations behind them
24:35 - Why these attacks are still human scale, and what changes if they’re not
26:45 - The fear: LLMs making hyper-personalized spear phishing easier at scale
29:25 - Money mule networks and why speed matters for remediation
34:25 - The “Swiss cheese model” and why BeeSafe advocates for multiple layers of defense
36:30 - Authorized push payment fraud, and the tension between usability and security
About Ariana Mirian
Ariana Mirian is a security researcher who has worked at the intersection of empirical measurement and security for over a decade, bringing a data driven lens to security decisions and intelligence. At BeeSafe AI, she is using this experience to drive novel methods that collect data on trust-based scams, in order to prevent victim losses. Prior to this, she worked as a senior security researcher at Censys, where she conducted measurement-driven research to map the Internet ecosystem. She earned her PhD in Computer Science and Engineering at UC San Diego, where she used Internet-scale measurement to study real-world security posture and develop data-driven methods across cybercrime, Internet-wide scanning, and enterprise security.
Connect with our guest Ariana Mirian: LinkedIn
Check out BeeSafe AI: https://beesafe.ai/
