An AI agent with access to crypto wallets got hacked this week through a prompt injection hidden in Morse code and routed through Grok.

It sounds funny. It’s also a preview of a security problem that Kain Warwick (Infinex/Synthetix) and Taylor Monahan (security expert, formerly of MetaMask) believe DeFi hasn’t seriously started to solve yet.

This segment breaks down how the attack worked, why agent-to-agent prompt injection is fundamentally different from smart contract exploits, and whether there’s any meaningful way to contain the blast radius once an AI agent has access to your keys.

Featuring RMIT research fellow Kelsie Nabben on the deeper structural questions behind AI security.

This clip comes from a longer conversation on the $71M KelpDAO funds court case, AI agent security, and Coinbase’s layoffs.

Full episode here: https://youtube.com/live/DXejCvRV-1s

We go live every Wednesday at 3:00 PM ET — subscribe to catch it live.

_______________________________________________________

Thank you to our sponsor!