A Conversation With Jonathan Kreiner @unsupervised-learning
A Conversation With Jonathan Kreiner  @unsupervised-learning
Uploaded November 2026 | Updated November 2026, 2 days ago
See how Orion Security stops data loss in real time with AI agents:
https://ul.live/orion-demo_yt

In this episode, I speak with Jonathan Kreiner from Orion Security about why data loss prevention has never really worked, and how AI agents that understand context, not just rules, can finally make it work.

We talk about:

• Why Rule-Based DLP Always Fell Behind
Traditional DLP starts with policies, and anything you didn't write a rule for walks out the door. Rules end up too broad (false positives) or too narrow (missed data), and the rule engine gets harder to run as the org grows.

• Context Over Anomalies
Why UEBA-style anomaly detection failed ("we're anomalous creatures"), and how Orion decides whether a piece of data moving at a given moment makes sense, using organizational context, data classification, and data lineage together.

• Data Lineage Without Containers
Instead of wrapping files in a container that breaks the moment someone copies, pastes, or screenshots, Orion follows the data itself, connecting downloads, uploads, renames, zips, and copy-pastes into one graph.

• Broad Telemetry From Day One
Endpoint sensors, a browser extension, direct SaaS and email gateway integrations, and enrichment from your IDP or Salesforce, plus webhooks for anything not covered yet.

• "The EDR for Data"
Doing the processing in the cloud instead of on the endpoint, scaling to tens of thousands of endpoints in year one, and alerts on the first day of a POV with no learning period.

• Orion as a Data Layer
Using the lineage graph through the web app, API, or MCP, including enriching EDR investigations ("what data did this malware touch?") and asking questions straight from Slack or Claude Code.

• DLP for AI Is Still DLP
How data exfiltration to ChatGPT and other AI tools is bringing CISOs back to DLP, and why allow-listing AI vendors isn't enough when personal accounts exist.

Chapters:
0:00 Intro: the problem Orion solves
0:44 Why rule-based DLP fell behind
2:09 Why anomaly detection failed
4:21 Organizational, data, and lineage context
5:20 Data lineage without containers
8:41 How Orion gets its telemetry
12:34 Why traditional DLP never delivered
15:01 Rolling out Orion
16:04 Adding business context and integrations
18:13 The interface: web app, API, and MCP
19:57 Customer wins at enterprise scale
22:11 "The EDR for data"
23:13 Asking questions from Slack and Claude Code
24:48 Resources and what's next
27:34 DLP for AI is still DLP

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 Jonathan KreinerI Summarized All Viktor Frankls Books Into 1 SentenceChinas New Way To Detect Stealth PlanesMy AI agents have emotions. I can hear when theyre excited.AI Is Getting Better At Brainstorming Than HumansUsing the Smartest AI to Rate Other AIRevealed! Putins Playbook To Take Down The WestAGIBOT Reveals Fleet of Humanoid RobotsWere All Building a Single Digital AssistantMastering Humanity: Your Future in the AI Era #shortsOpen AI Strawberry AI Is Scarily GoodHow and Why I Built PAI, with Nathan Labenz
Unsupervised Learning |

A Conversation With Jonathan Kreiner

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER