I Built an AI Agent That Doesn't Leak Sensitive Data
Source: YouTube · The AI Automators · published Feb 15, 2026 · 1:11:38
This video demonstrates a system that enables cloud LLMs to process private documents by automatically redacting and anonymizing sensitive data like PII before it leaves the user's network 0:00-0:13.
Key Takeaways:
• The architecture utilizes Microsoft Presidio for entity recognition and the Faker library to create realistic surrogate data, which generates better LLM responses than generic placeholders 5:44-6:08.
• Sensitive data is handled via two paths: hard redaction for irreversible items like Social Security numbers and reversible anonymization for names and locations 14:00-14:30.
• Complex entity resolution ensures that variations of a name (e.g., "Daniel Walsh" vs. "D. Walsh") map to the same surrogate to preserve semantic meaning for the AI 11:00-11:14.
• The solution was built using Claude Code's new agent teams feature, allowing parallel development of the ingestion pipeline, chat integration, and tool integration 19:00-19:20.
This technical safeguard addresses privacy regulations like GDPR but adds complexity in handling calculations involving dates or financial figures.
Sources:
- 0:00-0:13 Introduction to private document processing with cloud LLMs
- 5:44-6:08 Using Microsoft Presidio and Faker for realistic surrogates
- 11:00-11:14 Entity resolution for consistent mapping
- 14:00-14:30 Hard redaction versus reversible anonymization
- 19:00-19:20 Using Claude Code agent teams for implementation
Generate CPE Credits
Generate a professional CPE document from this video's transcript.
Estimated credit: 1 CPE hours
Estimate uses the video runtime (1 hour ≈ 1 CPE, rounded to the nearest 0.5, minimum 0.5, maximum 2.0). The final amount can be lower after review, never higher.
Topic: Cybersecurity. Commonly maps to: Security and Risk Management, Security Operations. Exact CISSP domains are assigned during generation.
Free account. One generation at a time, with a daily limit.
CPEBuddy is independent and not affiliated with or endorsed by ISC2, ISACA, or any certification body. Exports are formatted for common CPE submissions; acceptance is at your certification body's discretion.
Transcript Preview
First 800 characters of the transcript
What if you could use the most powerful cloud-based AI models with your private company documents and not have a single piece of sensitive information ever leave your network? No real names, no financials. The cloud LLM never sees any of it. Let me show you what that looks like. So, if we jump into chat and let's ask the question, uh, tell me about Sarah Williams who is a staff member. So, it's carrying out a hybrid search on Sarah Williams and we are getting back some information and this is all normal. You would expect this from an LLM. Let's say it's able to search documents and retrieve information. But what's interesting is that we're using Claude Haiku here and Haiku has not seen any of this information. So if we jump into Langfuse into this particular trace, first things first, you …