BeeSafe AI Deploys Undercover AI Agents to Waste Scammers' Time

The Macro: Scammers Are Using AI Too, and They Are Winning

Social engineering fraud is accelerating. Pig butchering scams, romance fraud, investment schemes, employment scams. The attackers have moved from crude phishing emails to sophisticated, long-term manipulation campaigns that unfold over weeks or months across SMS, voice, email, and social platforms. And they are using AI to scale these operations.

The numbers are staggering. Americans lost over $10 billion to fraud last year. The average pig butchering victim loses $100,000 or more. And the scam infrastructure keeps getting more sophisticated. Scammers create fake identities, build trust over time, and operate inside encrypted channels where traditional security tools cannot see them.

The defense side has not kept up. Banks use rule-based fraud detection that catches known patterns but misses novel attacks. Telecom companies can block known scam numbers, but attackers rotate numbers constantly. Law enforcement is overwhelmed. The fundamental asymmetry is that defense is reactive and scammers are proactive.

What if you could flip that asymmetry? What if instead of waiting for scams to reach victims, you went after the scammers directly?

The Micro: PhD Security Researchers Playing Offense

Ariana Mirian, Daniel Spokoyny, and Nikolai Vogler founded BeeSafe. Ariana has over ten years of experience as a security research scientist, previously at Censys and UCSD. Daniel has a PhD from CMU in machine learning and NLP with 10+ years in ML and security. Nikolai has a CS PhD from UCSD and was a CMU LTI researcher. This is a deeply credentialed team with exactly the right combination of security expertise and ML capability.

BeeSafe deploys undercover AI agents that engage with scammers directly. These agents respond to scam communications, play along with the manipulation, and use the interaction to identify the mule accounts and criminal infrastructure behind the scam. The agents waste scammers’ time and resources while collecting intelligence that financial institutions can use to block fraudulent accounts.

The approach is genuinely novel. Most fraud prevention is defensive. BeeSafe is offensive. Instead of building better walls, they are sending agents behind enemy lines. The intelligence they collect, especially the identification of mule accounts used to move stolen money, is extremely valuable to banks and payment processors.

Their backing includes YC, Obvious Ventures, and America’s Seed Fund through SBIR/STTR. That government research funding suggests the technology has passed serious technical review.

The Verdict

BeeSafe is taking a creative approach to a massive problem. Offensive fraud prevention through AI agent engagement is the kind of asymmetric strategy that could actually work. Scammers scale by running the same playbook against thousands of targets. If AI agents can consume their resources at scale, the economics of scamming get worse.

The risk is an arms race. Scammers will adapt when they realize they are being engaged by AI counter-agents. The question is whether BeeSafe can evolve faster than the scammers can adapt. The PhD-level ML expertise on the team suggests they can, but it will be a constant battle.

In 30 days, I want to see the number of mule accounts identified. That is the concrete output that financial institutions care about. In 60 days, the question is whether any bank has taken action based on BeeSafe intelligence. In 90 days, I want to know about the scale of scammer engagement. How many simultaneous scam conversations can BeeSafe’s agents maintain? The more they can handle, the more scammer resources they consume.