A Conversation With Harish Peri @unsupervised-learning
A Conversation With Harish Peri  @unsupervised-learning
Uploaded August 2026 | Updated September 2026, 3 weeks ago
Check out Okta here: https://ul.live/okta_yt

I had a great chat with Harish about the wild west of AI agents and why securing their access is basically just classic identity security applied to a brand-new surface. We dug into how things like Okta's agent gateway act as a much-needed bouncer for these bots, letting devs move fast without accidentally handing an AI the keys to the entire kingdom.

What we talk about:

The AI Agent Explosion: The rapid deployment of AI agents across enterprises and why their unchecked access to sensitive production data is keeping security teams up at night.

Back to Basics: Why the solution to securing AI agents isn't actually new, but rather a return to classic identity, authorization, and security fundamentals.

The Auditing Nightmare: The impending compliance challenges of proving the exact chain of custody from a user's initial prompt all the way to an autonomous agent's specific transaction.

Centralized Control and Kill Switches: How tools like Okta's agent gateway act as an independent control plane to manage resource connections, issue tokens, and execute an organizational kill switch if things go wrong.

Friction vs. Automation: The emerging threat of agent-to-agent scope escalation and why heavily relying on "human-in-the-loop" safeguards completely defeats the speed and purpose of AI

Here is a concise breakdown of the major questions and points raised by the interviewer:

00:00 - Intro and the core problem keeping customers awake.
04:30 - How to secure agents without a unified industry protocol.
08:22 - The "Mugging Face" incident and the need for immediate control.
16:15 - How Okta for Agents and the gateway technically operate.
18:50 - What installation looks like and how long it takes.
22:24 - Customer response to getting control over their agents.
26:28 - The Excel analogy and why "human-in-the-loop" defeats the purpose of AI.

Subscribe to the newsletter at:
danielmiessler.com/subscribe

Join the UL community at:
danielmiessler.com/upgrade

Follow on X:
https://x.com/danielmiessler

Follow on LinkedIn:
linkedin.com/in/danielmiessler
A Conversation With Harish PeriAIMM: A model for thinking about AI and WorkChatGPT Can Now Read Code From MacOS AppsMy New Favorite Fabric Pattern - extract_primary_problemHow To Actually Solve Bullying in SchoolA Deepdive on my Personal AI Infrastructure (PAI v2.0, December 2025)A Conversation with Jiquan Ngiam About Agent + MCP SecurityThe US Secretly Relies On China For BuildingAttackers vs. Defenders: The AI Stack Game #shortsAI Is Definitely IntelligentPredator Spyware Is Now Harder To TrackA Conversation with Shiladitya Sircar from BlackBerry on DeepFake Threats
Unsupervised Learning |

A Conversation With Harish Peri

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER