scripod.com
Building an AI Guardian for Enterprise with Onyx Security CEO Maxim Bar Kogan

Overview

Shownote

Highlights

Transcript

Chapters

Pins

Building an AI Guardian for Enterprise with Onyx Security CEO Maxim Bar Kogan

No Priors: Artificial Intelligence | Technology | Startups

May 28
No Priors: Artificial Intelligence | Technology | Startups

No Priors: Artificial Intelligence | Technology | Startups

May 28
OverviewShownoteHighlightsTranscriptChaptersPins

Shownote

We are now closer than ever before to living in a world where AI agents are smart enough to run our power grids and manage water supplies. How do we keep them from going rogue? Sarah Guo sits down with Maxim Bar Kogan, founder and CEO of Onyx Securities, t...

Highlights

This podcast explores the challenges and solutions for securing autonomous AI agents in enterprise environments. The conversation focuses on the need for specialized oversight systems that can monitor and control agent actions, moving beyond traditional security tools.
00:00
Enterprises face growing risks from AI agent actions.
00:45
Building AI agents to monitor other AI agents
04:03
Market for autonomous agents emerged with reasoning models.
05:17
It trains models and builds agents to oversee other agents.
07:52
Autonomous coding agents are the fastest growing category.
09:58
Existing tools lack context to understand agent intent
12:55
Understanding AI intent is the core challenge
14:11
Small models flag when a smarter agent needs to intervene.
18:40
Controlling advanced AI requires a vendor-independent oversight company.
21:25
Understanding internal weights will be part of the solution
23:35
Intense pain from AI security risks drives trust in early-stage solutions.
25:10
Long-term security requires foundational measures
27:45
Early AI release risks catastrophic infrastructure failure
30:00
Companies allowing flexibility will succeed.
30:46
Core pillars of LLMs and agent architectures have remained stable.
32:36
Enterprises need independent oversight for AI agents.
36:56
Israel has a deep adversarial thinking bench
39:22
Design for humans now, agents later.

Chapters

Cold Open
00:00
Maxim Bar Kogan Introduction
00:45
AutoGPT and Betting on Agent Actions
01:10
What Onyx Product Does
05:17
State of Deployment in Large Enterprises
07:47
Securing Agents
09:58
Why Proxies Don’t Work
12:45
Why Onyx Trains Its Own Models
14:11
Onyx’s Talent Culture
18:38
Mechanistic Interpretability
21:24
How Onyx Builds Customer Trust
23:35
Mitigating Risk at the Foundational Level
25:10
Phased Rollout of Glasswing and Daybreak
27:45
Large Enterprise Holdouts
29:11
Onyx and the Larger AI Security Space
30:46
Should Labs Address Model Trust and Governance?
32:36
What Needs to Happen in Security
36:56
Why Maxim is AGI-Pilled
39:14

Transcript

Maxim Bar Kogan: As you're exponentially doing more things with AIs, you're going to start having really bad actions happen. And we've seen some of that happen lately, with agents accidentally publishing code and tokens that they weren't supposed to. Like,...