Scuttlebutt

Scuttlebutt

AI will find security bugs faster than humans can patch them, says NEAR's co-founder

7 calls: why AI agents can fix a $20T economy of inefficient markets, the "uncensored" models NEAR built for defensive cybersecurity, and why NEAR just cut its own token issuance

Scuttlebutt's avatar
Scuttlebutt
Nov 07, 2026
∙ Paid

Milk Road · Nov 6, 2026 · 55m · 47m saved

With Illia Polosukhin, Co-Founder of NEAR Protocol and CEO of NEAR AI

AI will find security bugs faster than humans can patch them, says NEAR's co-founder

Illia Polosukhin, a co-author of the "Attention Is All You Need" paper that introduced the Transformer architecture behind modern AI, told Milk Road's John that he started warning of a "security Armageddon" after watching an AI model find a networking vulnerability that was nowhere in its training data. His answer is a stack that pairs AI agents with blockchain: agents that can trade, pay and now help secure software, built on a network he calls a "market of markets."

"The way I describe it is it's a vertically integrated stack for the future of AI and blockchain... everyone will have their personal or corporate agent that represents them." — Illia Polosukhin

The full interview is covered here so you can skip it. 55 minutes of audio, 8 minutes of reading.

Here are the 7 calls that matter.


This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.


Key Takeaways

  • AI agents can fix markets that run on guesswork, not data, Illia argues

  • NEAR went "confidential by default" in October, hiding balances and trades unless a user discloses them

  • Tireless AI will out-pace human bug-fixing, so NEAR built "uncensored" models just for defensive security work

  • A theorem-proving harness cut formal-verification costs roughly 200x, NEAR says

  • NEAR's Shield system paused a live exploit and clawed back stolen funds, after an October hack

  • NEAR Intents can't be fully permissionless, because it has to reconcile non-final blockchains

  • NEAR holders voted to cut annual token issuance from 2.5% to 1.6% over two years

1. Bad Data Makes Bad Markets

Illia's starting claim is that ordinary commerce runs on bad information, and that AI agents can close the gap. He points to San Francisco grocery stores, which he says overstock food by 50-100% because they have no real-time read on demand.

Illia says AI agents can make a $20T economy of guesswork efficient

"The stores in San Francisco stock up on food 50 to 100% more than actually gets bought... because there's no efficient information about how much purchasing will happen." — Illia Polosukhin

On near.com, that thesis is already live: users can swap Tesla stock or pre-IPO names for crypto, trade perpetuals, or route a payment to a European bank account, with execution against Nasdaq liquidity behind the scenes.

"Near.com already is effectively an exchange of exchanges — you can go from Tesla to Bitcoin to Zcash to someone's IBAN on a bank account in Europe." — Illia Polosukhin

He says the same model now reaches past financial assets into services and physical goods. Blockchain strips out the legal, banking and interchange costs layered across the roughly $20 trillion commercial economy — costs humans struggle to navigate, but an AI agent can route around them by turning plain language into on-chain commands.

2. Confidential By Default

User's avatar

Continue reading this post for free, courtesy of Scuttlebutt.

Or purchase a paid subscription.
© 2026 The Golden Age LLC · Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for writing, podcasting, and media founders.