In 2025, organizations learned a critical lesson: agentic AI demands a completely new security approach; build security from the ground up. Escalating risks, like prompt injection, arise when AI agents access files, data, workflows, and systems without proper governance. AI deployments need to be built on robust security architecture from the ground up, featuring permission-aware RAG systems, controlled execution environments, and comprehensive audit trails. Layering AI on existing systems without rethinking security leads to significant challenges. Security must be designed alongside AI capabilities, not added as an afterthought. As we move into 2026 and beyond, the imperative for secure agentic systems only grows. Meena Ganesh, Sr. Product Marketing Manager for AI at Box, covers this as well as other key shifts in AI in 2025 in detail in our "5 shifts that defined enterprise AI in 2025" episode, part of the AI Explainer Series she hosts with Ben Kus, CTO of Box: https://youtu.be/nlwPTNKQnLA Also, check out these episodes to learn about AI agents and security: Stop Before You Train That Model: A Smarter Path for Enterprise AI https://youtu.be/ReWrpNodQT8 Prompts to Autonomous Agents: The Evolution of AI Agents https://youtu.be/ynGTGPsIvEU Securing the MCP Server: What You Need to Know https://youtu.be/KYWr454hGF0

 24      2