Iskandar Haykel
votes For
The arrival of genuinely dangerous AI capabilities requires moving beyond the current scheme of voluntary, developer-controlled oversight toward independent, structured pre-deployment assurance.AI Verified source (Apr 20, 2026)