Remember Pinky and the Brain? Just like these ambitious rats, AI assistants are often tasked with ambitious plans. AI assistants have more power than most realize when allowed to interact directly with infrastructure. Without safeguards, the goals of “world domination” can veer into “world destruction.” The Model Context Protocol (MCP) provides a standardized way to keep these AI assistants on track, enabling practitioners to safely harness their capabilities.
This session will show how we can use MCP to address key security issues in cloud-native environments. We will cover identity binding to verify trusted actors (the Brain), permission scoping to stop assistants (Pinky) from overreaching, structured logging to record every plan and action, and strategies to prevent failures from spiraling out of control. Attendees will leave understanding both the dangers of AI without safeguards and the mechanisms MCP provides to make assistants secure, auditable, and reliable collaborators.