Vitalik's Privacy Manifesto, Illia's Bold AI Prediction, & Why NEAR Connects Both

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:00

   Good morning, Legion. Some big stories today, and they all come back to one key theme. Privacy and AI. First up, illia sat down with the Sequence for a long form interview covering Transformers, Private AI, Ironclaw and what comes next for Near.

2. 0:15

   On Confidential Compute, he explains that Near AI started with confidential inference because, quote, it's the most direct and has immediate value for customers from enterprises to privacy focused consumers. He also makes the point that their platform can host closed wait models where neither the hardware provider nor the consumer can access the waits, while model builders don't get access to consumer data.

3. 0:38

   On Agents, he has this to say, quote, The harness is as important as the model, pointing to examples where just changing how a smaller model edits files improves coding performance by a factor of 10. Ironclaw is built around that idea. Open source, written in Rust, secured from the ground up. Give it a try today and tell us what you think about it on the newly launched Ironclaw Reddit.

4. 0:59

   He also drops a prediction most people would disagree with. He says, quote, I think we'll see governments making nationalization proposals for AI labs. The only way to avoid that is if we have insured privacy of AI and provide decentralized access.

5. 1:12

   And last but not least, on commerce, one prediction he makes is that in one year agents are going to be facilitating global trade and paying via stablecoins. The full interview is worth a read. Check it out on the sequence. In our second story today, Vitalik just published a blog post on building a private self-sovereign AI setup.

6. 1:32

   Surprise, surprise. Local inference, local files, full sandboxing. He's direct about the problem, writing that, quote, we're on the verge of taking 10 steps backwards by normalizing feeding your entire life to cloud-based AI. Funnily enough, he sounds like illia two years ago. He calls out security issues with OpenClaw, including that roughly 15% of reviewed skills contain malicious instructions.

7. 1:54

   For remote inference, Vitalik says the best current option is TE-based compute, where you verify the hardware is only running a program that decrypts, processes, and encrypts with no logging. You can actually find a chapter of a talk given by Alex Skidanov on this exact topic from a couple of years ago.

8. 2:12

   Near AI Cloud runs exactly the setup in production today. Intel TDX, NVIDIA H200 TEEs, TLS terminating inside the secure enclave, and a full verification API for model and gateway attestation. What Vitalik describes as the ideal architecture, Near AI already has live.

9. 2:30

   If you haven't tried Ironclaw or NearEye Cloud yet, go to docs.ironclaw.com or cloud.near.ai and get started today! One quick ecosystem note to close out, the Drift protocol vault was recently drained. An on-chain analysis showed the attacker's wallet was initially funded through NearIntents before routing funds through intermediate wallets and bridging to Ethereum via Wormhole.

10. 2:51

    We don't condone any exploits, but it does show that even sophisticated actors are choosing Intents for privacy and speed. That's all for today. We'll do this all again real soon.