Home / Founders / Directory / Adam Wenchel
Adam Wenchel
Co-founder and CEO · Arthur
AI monitoring AI governance
Wenchel deployed AI for credit and fraud decisions at Capital One with no way to watch it in production, and started Arthur to give companies that view.
Adam Wenchel is co-founder and chief executive of Arthur, an AI performance company whose software monitors models and AI agents after they go live. He has worked in AI since the late 1990s, when he did research at DARPA while finishing a computer science degree at the University of Maryland, and he sold his first startup, Anax Security, to Capital One in 2015.
From an Atari 800 to DARPA
He started on home computers. “I used to program this Atari 800 computer back in the day, and I was really into games, so that’s kind of where I got started,” he told the University of Maryland before giving its 2025 science commencement keynote. As a freshman he helped set up one of the first streaming radio stations at Maryland. “Setting up our first server, I found myself staying up till 2 a.m. to work on it because I was just so excited,” he said. As a senior he followed his professor James Hendler, who had become a DARPA program manager, to do research there.
Anax Security and Capital One
Anax Security used AI to detect and block cyberattacks. After Capital One bought it, Wenchel worked with the bank’s chief executive and chief information officer to start its Center for Machine Learning and grew it to nearly 300 people as vice president of AI and data innovation, he told Work-Bench. The models went into credit decisions, cybersecurity and fraud, and there were no tools to monitor or explain them, which he said “kept me awake at night.”
Founding Arthur
He started Arthur in 2019. “What we focus on is, if you’re putting AI systems out in the world, how do you monitor them to make sure they’re saying the right thing and making good decisions, and if they’re not, we provide insight into why that is,” he told the university. In 2023 the company launched Arthur Shield, a firewall for large language models that screens for data leaks, hallucinations and prompt injection.
