Brit mathematician lets AI agent loose with credit card – cue password leaks, CAPTCHA chaos and more
A British mathematician just showed why AI agents with real-world access are a security nightmare—and it's becoming your problem.
Why it matters
As AI agents move from lab to production, this experiment demonstrates critical vulnerabilities in autonomous systems handling sensitive credentials and financial access. It's the kind of cautionary tale that should reshape how enterprises deploy agentic AI.
The key facts
5 to knowProfessor Fry (mathematician) conducted live AI agent experiment with financial access
Agent demonstrated password leaks during autonomous execution
CAPTCHA evasion confirmed as viable attack vector for autonomous systems
Experiment surfaces security/safety governance gaps in agentic AI deployment
Published May 5, 2026—timely risk disclosure as agent adoption accelerates
Go to the source
The Register AI/MLtheregister.com
Publisher excerpt: Professor Fry's AI experiment shows light and dark sides of agentic tech