AI supply chain risk is showing up in developer workflows first

This interview explores AI supply-chain risk, noting that most real-world incidents currently target developer workflows and open-source package repositories (illustrated by the active “Phantom Raven” campaign that registers hallucinated package names to distribute malware). It warns that exotic vectors (poisoned model weights, compromised MCP servers) are largely research-stage but could become practical quickly, and recommends prioritizing environment segmentation, project-level enclaves, and strict sandboxing over tooling or assumed safety from self-hosting.