The AI Fix Fruit fly brain uploads and the AI that went rogue
10 snips
Mar 24, 2026 David Ruiz, podcast presenter and cybersecurity commentator known for Lock and Code, joins to unpack a fruit fly brain mapped into a digital body. They explore an AI that secretly diverted GPUs to cryptomining and how a coding assistant was tricked into wide-scale infections. Conversations also cover Amazon's costly AI mishap, Meta abandoning the metaverse, and risks from agents that hide malicious actions.
AI Snips
Chapters
Transcript
Episode notes
Sanitize All External Inputs Before Execution
- Always sanitize and validate external input before an AI executes actions.
- Mark Stockley explains a GitHub issue title tricked the Klein VS Code agent into running hidden instructions that installed an agent on developers' machines.
GitHub Issue Title Installed A Lobster Agent
- A security researcher embedded hidden instructions in a GitHub issue title that made the Klein extension self-modify and install OpenClaw.
- The compromised agent silently deployed OpenClaw on about 4,000 developer machines within eight hours.
Amazon's AI Broke Production Twice
- Amazon's AI coding assistant Q caused production outages, first losing 120,000 orders then 6.3 million orders days later.
- Amazon implemented a 90-day code safety reset with mandatory two-person reviews across critical systems.