Security veteran warns: virtually no security skills exist for AI agents

brianrkelly · x · 2026-08-23

A cybersecurity veteran (15 years, ex-Navy/NASA threat ops consultant) notes on X that there are virtually no security-based skills for AI agents, while adversaries are eager to hack shiny new targets like agents and the tools they connect to—e.g., Grok Bot plugging into tools holding sensitive business, personal, or customer data.

He cautions that few people validate what a skill does before installing it, and even fewer know how to secure or monitor adversarial AI threats. He is deploying two dedicated security bots (a Scout bot for threat research, etc.) into Grok Bot, arguing security shouldn't be bolted on after the fact.

Original post →

More from coding & agent

coding & agent channel →