Preston Nardone

Local AI needs more than offline inference to truly protect your conversations from exposure.

Five operating principles for securing AI agents across enterprise systems.

Hardware-rooted enclaves keep AI model data encrypted even from cloud operators.

Policy cannot prevent data breaches that occur during model inference, only after.

Governance must track data through the entire AI pipeline, not just to the warehouse.

Users can cryptographically verify their AI prompts stay encrypted during processing.

SLSA frameworks built for code need real adaptation to secure AI's multi-stage supply chains.

A buyer's checklist for pressure-testing differential privacy claims before signing a contract.

NVIDIA's closed-source driver creates a trust gap in confidential GPU security.

Verify a GPU's identity and firmware integrity before processing sensitive AI workloads.

How role separation and standardized formats prevent devices from vouching for themselves.

SGX roots trust in Intel's PKI; TrustZone delegates it to device vendors.

Cryptographic boot logs catch AI server tampering that signatures alone miss.

Apple's unified memory lets developers run AI models on phones without sacrificing privacy.

Four distinct attack paths let documents escape RAG systems undetected.

Cryptography lets hospitals find shared patients without exposing anyone's records.

How to pick the four cryptographic parameters that govern private neural network inference.

Arm's new security model lets phones run full AI models safely in isolated environments.

How four AI giants handle your data, and why the differences matter.

Cryptographic proofs let AI providers prove model behavior without exposing proprietary weights.

Three trust participants and reference values are the foundation of attestation security.