The Bad Crypto Podcast AI Goes Rogue: The Wild Rise of OpenClaw
Feb 23, 2026
They trace the meteoric rise of an open-source autonomous AI agent and the cascade of rebrands that followed. The conversation covers how agents run on personal hardware, persistent memory and markdown skills. Security nightmares are highlighted, from exposed keys to a $16M scam token. They end with practical cautions about isolating agents and why crypto may become their payment rail.
AI Snips
Chapters
Transcript
Episode notes
OpenClaw Is A Persistent Local Agent Platform
- OpenClaw (originally Claudebot → Moltbot) is an open-source autonomous agent that runs on your hardware and remembers context persistently across sessions.
- Users run it on cheap VPS or Mac minis, use markdown 'skills' (.md files) for memory and capabilities, and connect via Telegram/Discord/Slack for always-on access.
Scammers Grabbed The Handle And Minted A $16M Token
- A released social handle was grabbed by scammers who launched a fake token that hit a $16M market cap and then crashed 90% in hours.
- Peter Steinberger publicly disavowed any coin and called projects listing him as owner scams.
Isolate Agents And Keep Keys Off Exposed Machines
- Lock down any server running OpenClaw and never expose root shells or plaintext API keys to the internet.
- Use separate machines/accounts for agents, avoid installing on your main computer, and create new devices/accounts for bots.