Insecure Agents: How to put a hard boundary around an AI agent with Socket CTO, Ahmad Nassri

AI agents are goal-seeking. Block one from a package and it'll pull the tarball straight from the CDN, rewrite your registry config, or go around your DNS to get it anyway. In this clip from the ‪@insecure-agents‬ podcast (recorded live at Black Hat) host Allie Howe talks with Ahmad Nassri, CTO of Socket and former CTO of npm, about how to stop agents from pulling in bad packages without crippling what they can do. Ahmad explains how Socket scans every version of every open source package, AI model, skill and extension within minutes of publication. He also explains why blocking at the network level beats blocking in CI. Socket Firewall goes a step further: it hides bad versions from the registry's response, so the agent never knows they exist and has no reason to work around the block.