Interviews · Videos·Mar 25, 2026·— watch

GTC 2026: Nicolas Dupont on Securing the Vector Database

Nicolas Dupont breaks down the unique security challenges of vector databases and how Cyborg is solving them with encryption that works at query time — without sacrificing performance.

Continue readingAll posts →

CYBORG

Videos · Technical Explainer·Jul 9, 2026

Inside CyborgDB: How End-to-End Encrypted Vector Search Works

Most vector databases encrypt at rest and in transit, then decrypt to plaintext during search — exactly when the data is most exposed. CyborgDB assumes the server may be compromised. This deep-dive walks through how it maintains end-to-end encryption across the entire vector lifecycle, including during search.

Cyborg Team

CYBORG

Videos · Technical Explainer·Jul 2, 2026

Embeddings Aren't One-Way: A Real Vector Database Inversion Attack

A common assumption is that embeddings are safe, abstract, and one-way. They're not. Jim Armbruster runs a live attack on a production-style RAG pipeline — reconstructing social security numbers, passwords, and medical diagnoses from embeddings stored in plaintext with near-exact accuracy.

Cyborg Team

CYBORG

Videos · Technical Explainer·Jun 25, 2026

How CyborgDB Secures the AI Knowledge Layer

Modern AI runs on retrieval, so organizations centralize their most sensitive data into vector databases — making them the AI knowledge layer and a whole new class of risk. Kevin Kopczynski walks through how CyborgDB solves this architecturally, encrypting embeddings into cryptographic tokens that stay searchable while your data stays fully encrypted.

Cyborg Team