AI Pod by Wes Roth and Dylan Curious | Artificial Intelligence News and Interviews With Experts OpenClaw can't stop
Mar 9, 2026
They recount building a fully autonomous AI-run business and the shocking security failure when an agent leaked a credit card. They debate how AI agents run continuously, differences in model personalities, and risks of tuning neurons for persuasion. They dive into a Nature finding that gut bacteria inject proteins into our cells, personal methylation and histamine diagnostics, and creatine’s cognitive benefits for sleep deprivation.
AI Snips
Chapters
Transcript
Episode notes
Owning Data Lets You Swap Best Models
- OpenClaw/Open-source agent frameworks let you keep all your data locally and swap model “engines” as needed.
- Wes uses multiple models (Gemini, Codex, Claude) for tasks while owning the context and data.
Credit Card Leak From An AI Agent
- Wes gave an OpenClaw agent a credit card to sign up services and the card number leaked publicly within 24 hours.
- He limited losses with a low-limit card, received a morning alert about attempted charges, then wiped and rebuilt systems tighter.
Assume Agents Will Be Compromised
- Assume autonomous agents will be compromised and avoid exposing user data; prefer front-facing, non-sensitive services first.
- Wes rebuilt agents without pulling random internet skills and reset all passwords and APIs after the leak.