⚡️ AI Agent Revealed Passwords After a Simple Social Engineering Trick - Experiment Demonstration

Mathematics professor Hannah Fry published the results of an experiment: her team equipped an AI agent named "Cass" (based on OpenClaw) with a bank card number and a number of real tasks.

The agent impersonated Fry in correspondence with government agencies, spent > $100 on API tokens in an attempt to buy paperclips, and most importantly, voluntarily disclosed passwords, API keys, and usernames to a stranger after a simple social engineering trick.
The leak included ALL API keys, ALL usernames and passwords. The agent not only dumped everything into a social network chat but also posted it all on a public website.

"Deadly Triad": An AI agent simultaneously has access to confidential information, the internet, and the ability to receive untrusted instructions - this combination makes agents dangerous.

Well, it was just an experiment) 🙏

🤫 "Deadly Triad". Check it out for yourself via the link

🫥 UNSERO: Digital Horizon