Agentic AI represents a fundamental regression in security, and an amplification of a 60 year old problem... and no one is talking about it.
https://anthonymccloskey.com/architecture-of-a-nightmare
We’re seeing a gold rush toward Agentic AI; tools that don't just chat, but actually act in our IDEs, our browsers, and our terminal.
But with 20+ years in cybersecurity experience, from the U.S. Navy to Comcast, Twitter, and Salesforce, I see a familiar, dangerous pattern repeating.
We are taking the "Original Sin" of computing, the inability to distinguish between data and instructions, and we are deliberately amplifying it. When an AI agent reads an untrusted email or a web page, that data becomes a command.
https://anthonymccloskey.com/architecture-of-a-nightmare
We’re seeing a gold rush toward Agentic AI; tools that don't just chat, but actually act in our IDEs, our browsers, and our terminal.
But with 20+ years in cybersecurity experience, from the U.S. Navy to Comcast, Twitter, and Salesforce, I see a familiar, dangerous pattern repeating.
We are taking the "Original Sin" of computing, the inability to distinguish between data and instructions, and we are deliberately amplifying it. When an AI agent reads an untrusted email or a web page, that data becomes a command.
