#745: The AI Approval Layer Is Fake with Zach Herbert

Zach Herbert, co-founder of Foundation Devices, explains how the company applies Bitcoin security principles—such as air-gapping and explicit human approvals—to address AI's security risks. Originally focused on Bitcoin-only hardware wallets like the Passport series, Foundation has developed Passport Prime and its microkernel operating system, KeyOS, to secure broader digital life amid AI's rapid integration.

Bitcoin Principles for AI Containment

Foundation views AI as a powerful tool that outpaces human oversight in computing tasks, yet current usage grants it unchecked access. Herbert notes that AI models like Claude or Cursor request "approvals" on the same device they control, creating a fake permission layer: once connected to tools like Gmail or APIs, they have full capabilities and merely ask politely. This mirrors risks Bitcoiners avoid through hardware wallets, where humans review transactions on trusted, isolated devices.

To contain AI, Foundation advocates separating execution from approvals. Routine actions (e.g., paying familiar invoices) could proceed autonomously, but high-stakes ones (e.g., Bitcoin transfers) require human sign-off on air-gapped hardware. This leverages AI for speed—Foundation's small team adds features rapidly—without surrendering control, avoiding the need for massive hiring.