BeeSafe AI
ActiveFrontier AI Defenses for Social Engineering Attacks
About
BeeSafe AI is building the frontier model layer for social engineering defense. Cyberattacks have moved from code to conversation: attackers now use AI to carry out long-term, trust-based manipulation across SMS, voice, email, and social platforms, often inside encrypted channels where defenders cannot see, detect, or stop them. BeeSafe changes that by deploying undercover AI agents that engage attackers directly, expose hidden infrastructure, and convert live adversarial conversations into actionable intelligence. Each engagement adds to BeeSafe's proprietary dataset of real attacker behavior and trains foundational models that continually learn about counter-persuasion, deception, and manipulation in the wild. BeeSafe is starting in financial crime, where banks already value our data against Authorized Push Payment (APP) fraud and money laundering, but the opportunity is far broader. The same data and models that uncover mule networks today can protect consumers, secure enterprises, and strengthen governments against the next generation of AI-powered social engineering attacks tomorrow.
Founders · 3
I am a machine learning and security researcher with 10+ years in ML and NLP. I earned my PhD at CMU with research on novel transformer architectures, better training objectives, and evaluation benchmarks for reasoning.
Cofounder at BeeSafe AI. I have over ten years of experience as a security research and measurement scientist, using large scale data to make better security decisions. I've worked in various organizations, such as Censys, Google Chrome, and UCSD, all of which have provided unique perspectives into security at scale to protect the end user. My goal is to make the Internet a safer place for everyone, regardless of background or technical expertise.
CS PhD from UCSD and ex-CMU LTI. My objective is to reduce human-driven cyber risk after hearing so many stories from peers about how they’ve gotten scammed. I have worked in machine translation, optical character recognition, and document attribution, which was actually interrupted by a real-world ransomware attack.
Related startups
Red teaming for AI agents
We build networks of AI personas that simulate stakeholder opinions
