AI agents expose the security checks you never actually wrote
How attackers took twenty thousand Instagram accounts by asking Meta's AI politely, and why that failure is about to become common.
Fabio Salvadori
AI systems architect and indie builder focused on agentic AI, trustworthy automation, and developer infrastructure. I’m the creator of PIC Standard, an open-source governance layer for high-impact AI agent actions, and NCP, a protocol-oriented approach to deterministic and replayable agent evaluation. My work focuses on making AI agents safer to operate in real software environments: clearer intent, stronger constraints, better provenance, approvals, receipts, and debuggable execution.
