*� NEAR’s Illia Polosukhin Has a Plan for Open-Source AI to Beat the Labs

From Transformers to encrypted trading bots — NEAR’s co-founder is rewriting the AI endgame.

*� Key Highlights

• 👨‍🔬 Illia Polosukhin (co-author of “Attention is All You Need”, co-founder of NEAR) is now building NEAR AI
• 🛡️ Focus: user-owned, privacy-preserving agents powered by TEEs + smart contracts
• ⚙️ NEAR’s stack supports encrypted inference + on-chain coordination with AI agents
• 🏥 Early use cases: healthcare, finance, gov — sectors where privacy is non-negotiable
• ⏱️ Illia predicts open-source AI will outpace Big Tech labs in the next 18–24 months

*� From Google to Open Infrastructure Illia Polosukhin spent years at Google pushing AI boundaries. Now he’s betting on decentralized, agentic AI.

“Imagine you’ve got 1,000 reports. As a human, you’d go crazy. But an AI? It scales.”

On Software Engineering Daily, Illia laid out NEAR’s roadmap for AI built on privacy, modularity, and ownership — where developers never touch user data, but can build world-class agents.

*� Trusted Execution Environments = Encrypted AI The core? Trusted Execution Environments (TEEs):

• Everything is encrypted end-to-end
• AI inference happens inside the Secure Enclave
• You get cryptographic proof (from Intel/NVIDIA) of the computation

No devs, cloud hosts, or middlemen can peek — and yet agents stay verifiable.

*� Real-World Use Cases Rolling Out

• 🏦 AI for finance that doesn’t leak trades or strategy
• 🧬 Medical inference without patient data leaving local hardware
• 🧑‍💼 Bots that transcribe or summarize — but never store or expose content
• 🤖 Smart contracts that interact with AI agents on-chain with full transparency

“You can build agents where even you don’t know what users are asking,” Illia says. That’s essential for regulated industries.

*� The NEAR AI Stack: Modular and Live

• 🧠 Inference Layer — call open AI models privately
• 👤 Agent Layer — run encrypted user-specific logic
• 📜 Protocol Layer — coordinate agent workflows on-chain
• 🛍️ Marketplace Layer — monetize, deploy, and compose over 1,000 agents

This opens the door for:

• AI-native credit scores
• DAO agents that audit proposals
• Autonomous marketplaces for compute, data, or labor

⏳ The Countdown to Open-Source Dominance Polosukhin’s take:

“In 18–24 months, open-source AI will move faster than the labs. If coordinated properly, open infra beats closed every time.”

Why?

• Hardware is catching up (Intel, AMD, NVIDIA all onboard)
• Open agents can self-improve via on-chain incentives
• Coordination across compute, research, and capital becomes native

NEAR’s bet: decentralized agent networks will win.

*� What Developers Can Build Right Now

• 🧪 Agents deployed via Docker
• 🧰 Integrated support for Intel, AMD, NVIDIA, Azure TEEs
• 🧾 Rust or JS smart contracts coordinating agent workflows
• 💼 Monetization tools + open agent registries

You don’t have to wait. NEAR’s privacy-preserving, AI-native stack is live.

*� AI That Belongs to Its Users NEAR AI isn’t just about infrastructure — it’s about flipping the power dynamic:

Agents that don’t leak. Compute that’s owned. Data that stays private.

Whether you’re building healthcare AI, trading bots, or LLM copilots — NEAR wants to give you the tools to own your future.

⚡ TL;DR Illia Polosukhin — co-author of the Transformer paper — is building out NEAR AI: a privacy-first, open-source stack for agentic AI. Encrypted computation, verifiable agents, on-chain coordination — all live now. In 2 years, he says, open AI infra will leave closed labs in the dust.