Bankless

Illia Polosukhin: Why AI Agents Are Still Useless (And What Fixes Them) | NEAR Founder on IronClaw

1h 18min

Illia Polosukhin, founder of NEAR, discusses why current AI agents lack utility, primarily due to critical security and privacy flaws where user secrets are exposed to LLM providers. He introduces IronClaw, a system designed to secure AI agents through defense-in-depth mechanisms and confidential cloud inference, enabling a future of self-sovereign AI. Polosukhin envisions AI as the new computing interface, with blockchain acting as its essential, trust-minimized backend for identity, markets, and governance, fostering a world where AI enhances human capabilities and sovereignty.

Summarized by Podsumo

🎙️ Ask about this episode

✨ Key Takeaways

1

Current AI agents are "useless" due to security and privacy risks: Tools like OpenClaw often send user secrets (API keys, tokens) to centralized LLM providers, making them untrustworthy for sensitive tasks.
2

IronClaw provides defense-in-depth security for AI agents: It vaults credentials, enforces policy-based actions, isolates tools in WebAssembly VMs, and detects prompt injection and data exfiltration, preventing agents from misusing or leaking sensitive data.
3

Confidential AI cloud enables private inference: NEAR AI Cloud offers a service running open-weight models in secure enclaves using multi-party computation (MPC), ensuring neither the model provider nor hardware provider can access user data or inference.
4

Blockchain as the backend for AI's future: Illia sees AI as the new operating system, with blockchain providing the "root of trust" for identities, marketplaces, micro-payments, and upgradable governance, bridging traditional and digital systems.
5

Autonomous agents and businesses are the future: Leveraging blockchain, agents can become self-determining, persistent, and even form "autonomous businesses" governed by tokens, promoting a "decentralized accelerationism" that enhances individual sovereignty.

💬 Notable Quotes

"So somewhere in an ontropic open AI log, they have everybody's access keys, API keys, and bear tokens to access your Gmail and your notions. It's actually insane that we're doing that."

— Illia Polosukhin
"I usually say AI is a user interface blockchain is a backend, right?"

— Illia Polosukhin
"The idea is like you should know what AI model you use you should be able to access system prompt and you should be able to... This and obviously most users will not do it but the experience should be very easy right and like people can inspect that indeed everything is straight forward and clear and it needs to be preserving your privacy, your data, your ownership over it."

— Illia Polosukhin