Season 2 · Episode 1235
Beyond "No Training": Securing the New Agentic AI Stack
Think your data is safe because of a "no training" clause? We deconstruct the hidden security risks within the modern agentic AI stack.
My Weird Prompts · Daniel Rosehill
March 15, 202630m 57s
Audio is streamed directly from the publisher (dts.podtrac.com) as published in their RSS feed. Play Podcasts does not host this file. Rights-holders can request removal through the copyright & takedown page.
Show Notes
As we move from simple chatbots to autonomous agents with long-term memory, the standard "we do not train on your data" marketing promise is no longer a sufficient guarantee of enterprise security. This episode deconstructs the "agentic stack," revealing how sensitive information flows through vector databases, orchestration layers, and observability tools that often lack the rigorous protections of the base model providers. By examining the technical shift from stateless interactions to stateful relationships, we uncover why your data is arguably more at risk in 2026 than ever before, while providing a concrete audit framework to help developers protect their infrastructure from leaks, vector inversion, and unauthorized access.