Karsten Wildberger
votes For
Require human-in-the-loop oversight for agentic AI systems acting in high-stakes domains
Safety and security cannot be added at the end. They must be built in from the beginning. From the start.AI Verified source (Jul 7, 2026)