Small Models Could Crack the Private AI Problem
Fully homomorphic encryption is years away from frontier scale. But the bar for "good enough" just got lower.
Two weeks ago we discussed Vitalik's updated framework for how Ethereum intersects with AI, which fittingly arrived as Anthropic's Opus 4.6 report showcased concerning emergent behaviors in testing. Then, and once again in yesterday’s premium article reiterating the synergy between AI and crypto, I stressed two pillars of Vitalik's analysis:
As we know, AI is becoming embedded everywhere as agentic abilities progress. Samsung just announced that Perplexity will be integrated into Galaxy AI on the S26 series with access to device context, following the trend of others like Apple Intelligence and Gemini Nano that are all moving AI closer to the user.
While on-device models are the final frontier approaching quickly, AI has moved closer to the user across all fronts. People give Claude, ChatGPT, and Perplexity access to their calendars, emails, and files every day through desktop apps and browser extensions, especially since the advent of Claude Cowork. Either way, AI systems are getting deeper access to personal data, whether they run locally or call home to a server.
Subscribe for free to continue reading
- Support the Bankless Movement
- Access to thousands of articles
- Complete archive of Bankless episodes
- Embark on free quests in Airdrop Hunter
- Daily alpha in your inbox
Already subscribed? Sign in
David Christopher
David is a writer/analyst at Bankless. Prior to joining Bankless, he worked for a series of early-stage crypto startups and on grants from the Ethereum, Solana, and Urbit Foundations. He graduated from Skidmore College in New York. He currently lives in the Midwest and enjoys NFTs, but no longer participates in them.
