The Rollup The Privacy Problem No One's Talking About in AI with George Zeng
Feb 6, 2026
George Zeng, Chief Product Officer at NEAR and GM for NEAR AI, builds privacy-preserving, decentralized AI tools. He talks about where crypto capital is moving, NEAR’s confidential inference and private chat using TEEs, the rise of OpenClaw agents and security tradeoffs, local-first AI benefits, and how NEAR Intents enable real-world agent actions.
AI Snips
Chapters
Books
Transcript
Episode notes
NEAR's AI-First Origin And Decentralization Goal
- NEAR was founded as an AI-first project and later built blockchain infrastructure to enable user-owned AI.
- The team aims to decentralize AI so it remains open, competitive, and not controlled by a few companies.
Host Models In TEEs For Confidential Inference
- Run models inside Trusted Execution Environments (TEEs) so inference data never leaves a secure enclave.
- Also evaluate application-level security and guard against prompt-injection vulnerabilities.
Quarantine Agents Off Your Primary Machine
- Do not run OpenClaw on your primary machine that stores passwords or financial data.
- Instead deploy the agent in a quarantined confidential VM or use a hosted service to limit exposure.
