Unchained The Chopping Block: AI's Role in Crypto, Agentic Coding, & Citrini Financial Crisis
40 snips
Feb 27, 2026 
Illia Polosukhin, co-founder of NEAR Protocol and contributing author of the original Transformers paper, joins to discuss AI meeting crypto. They explore agentic assistants as new interfaces. Topics include secure enclave hosting, permissioning lessons from blockchain, AI-written smart contract risks, agent marketplaces and escrows, and the debate over sensational AI-driven financial doom.
AI Snips
Chapters
Transcript
Episode notes
Agents Replace Apps As The Interface To Intent
- Agentic assistants (Claws) change computing by executing multi-step real-world tasks like scheduling, shopping, and trading on your behalf.
- OpenClaw and IronClaw show the shift from single-query chatbots to looped agents that act on credentials and APIs.
OpenClaw Failures Highlight Agent Dangers
- Real OpenClaw incidents exposed severe risks: one user had hundreds of emails deleted after compaction and another agent accidentally issued a $400k-equivalent token transfer.
- These stories show how context overflow and poor permissioning produce catastrophic agent behavior.
Design Agents With Blockchain Threat Models
- Build agents with blockchain-grade threat models: assume malicious contributors, supply-chain attacks, and continuous adversaries.
- Near.ai uses TEEs, WASM, MPC key-holding, encrypted credential injection and policy-based spend limits to limit damage.