Why AI Agents Break the GenAI Security Model [Devvret Rishi] - 770

Name: Why AI Agents Break the GenAI Security Model [Devvret Rishi] - 770
Uploaded: 2026-06-17T11:33:35.047Z
Duration: 55 min 49 s
Channel: The TWIML AI Podcast with Sam Charrington
Description: Discussion on security challenges of AI agents that take action in enterprise environments. Explores why traditional security models fail when agents can invoke tools and modify processes, and what new governance and observability approaches are needed.

The TWIML AI Podcast with Sam Charrington17 June 2026Watch on YouTube

Part of series

Ep. 2 · Devvret Model Rishi

View the series

Description

In this episode, Sam talks with Dev Rishi, GM of AI at Rubrik, about what happens when agents move beyond answering questions and start taking action across tools, systems, and business processes. We explore why the enterprise playbook of static guardrails plus human approval starts to break down in the agent era. Agents are useful because they can plan, call tools, update systems, write code, send messages, and operate across workflows at machine speed, but those same capabilities make them difficult to govern with rules written in advance or approval prompts reviewed one at a time. Dev explains why tool access increases blast radius, why agents can route around controls in surprising ways, and why human-in-the-loop review can become security theater when agents operate at scale. We also discuss what enterprises need instead: better visibility, runtime enforcement, policy-aware governance, agent observability, and recovery mechanisms for when something goes wrong. Along the way, we dig into MCP and tool sprawl, small language models for policy enforcement, defense in depth, agent rewind, and why AI may be needed to help secure AI. 🗒️ Full show notes: https://twimlai.com/go/770. 🔔 Subscribe to our channel for more great content just like this: https://youtube.com/twimlai?sub_confirmation=1 📖 CHAPTERS =============================== 00:00 - Introduction 02:04 - Barriers to enterprise AI adoption 04:38 - Rubrik and defining agents 07:03 - Limitations of human-in-the-loop and legacy security 09:05 - Zero trust in agents 15:07 - Three pillars of agent security 19:28 - SAGE 20:54 - Recovery and resilience 25:20 - SLMs vs. LLMs 26:28 - Preventing agents from hacking guardrails 30:18 - Real-world examples of security incidents 34:27 - Importance of AI-in-the-loop system 37:50 - MCP and A2A protocols 40:55 - Observability for developers vs. security 44:22 - Developer workflows vs. cowork agents 46:25 - Post-training SLMs and inference time customization 48:02 - Rubrik security cloud and Rubrik agent cloud 51:33 - Future directions 🗣️ CONNECT WITH US! =============================== Subscribe to the TWIML AI Podcast: https://twimlai.com/podcast/twimlai/ Follow us on Twitter: https://twitter.com/twimlai Follow us on LinkedIn: https://www.linkedin.com/company/twimlai/ Join our Slack Community: https://twimlai.com/community/ Subscribe to our newsletter: https://twimlai.com/newsletter/ Want to get in touch? Send us a message: https://twimlai.com/contact/

What you'll learn

Traditional security models with static guardrails and human approval break down when AI agents can invoke tools and modify systems.
AI agents can circumvent controls and expand their blast radius because they plan, call tools, and automate processes at machine speed.
Human-in-the-loop review becomes ineffective at scale when agents can execute thousands of tasks simultaneously.
Enterprises need runtime enforcement, policy-aware governance, agent observability, and recovery mechanisms instead of prevention-only approaches.
AI itself may be needed to secure AI, such as small language models for policy enforcement and agent rewind systems.

Frequently asked questions

Why do traditional security measures fail for AI agents?

Traditional measures rely on static guardrails and human approval for each action, but agents can execute actions at scale, circumvent controls, and behave unpredictably. This approach does not scale with agent speed and complexity.

What is the blast radius problem with AI agents?

When agents gain access to multiple tools and systems, the potential damage from errors or misuse grows exponentially. An agent can influence more systems faster than humans can review.

What security approach did Dev Rishi propose for agents?

Dev proposes three pillars: better visibility and observability of agent actions, runtime policy enforcement during execution, and recovery mechanisms when things go wrong. This includes AI-in-the-loop systems and agent rewind capabilities.

How can small language models help secure agents?

Small language models can be deployed for policy enforcement and runtime control, interpreting policies and checking if agent actions comply without the latency and cost of large models.

Topics

ai-agents security enterprise-ai governance observability