PodcastBanklessIllia Polosukhin

The Government Just Put a Border Around AI

Export controls on advanced models are already reshaping geopolitics, creating incentives for sovereign AI labs and practical pressure to collect KYC. Technical responses under discussion include decentralizing inference, building confidential, verifiable stacks, and agent marketplaces, while NEAR is positioning products—near.com and a private shard—to offer privacy and KYC‑resistant entry points. Market implications include faster adoption of lower‑cost, near‑frontier AI in emerging markets, rising confidential TVL on NEAR, and a bullish view on privacy‑oriented blockchains if regulatory frictions around banking and transfers tighten.

AI SafetyGeopolitical RiskArtificial Intelligence

Preview · 18 more points in the app

Read the Full Summary

Key Takeaways

Recent export‑control actions on advanced models, plus framing AI as a national‑security threat, create political pressure for sovereign labs and identity‑based restrictions (age/nationality), which in turn motivates decentralized inference and confidential, verifiable stacks so agents can run locally without government telemetry.

NEAR is positioning itself as a KYC‑resistant, privacy‑oriented stack: near.com offers passkey/Face ID UX and wallet consolidation, NEAR launched a private shard and confidential API with a cross‑asset 'privacy set,' confidential TVL is rising, and the guest argues the protocol looks especially attractive if banking and transfer restrictions tighten.

There is a persistent tension between centralized providers and privacy‑first alternatives: polished UX and KYC reduce user friction and favor incumbents, but vendor control (capability restrictions and changing data retention policies) can materially erode utility and drive demand for private, user‑owned solutions.

Enterprises in finance, legal and medical want privacy‑first, user‑owned AI because routing decisions through models increases the risk that providers could replicate firm IP; this concern is compounded by demonstrations that even a local 7B model can design dangerous biological constructs, though materializing those designs remains a separate challenge.

GEO-POLITICS

Export Controls — Recent export‑control moves on advanced models set a precedent for using export controls to restrict internet services more broadly. That creates practical walls between the United States and other countries and signals that services can be shut down.

“It's also now puts every single other government in the world on alert that at any time not just AI effectively every single internet service can be shut down.”—

Sovereign Labs — Framing AI as a national‑security threat gives governments a political rationale to build sovereign AI labs and to nationalize or tightly control AI providers. That pressure, paradoxically, motivates decentralized or cross‑border development.

“I think like models are coming I think the expert control is honestly was more of a like a really bad precedent that now I'm going to activate actually even…”—

TECHNOLOGY

Decentralized Inference — Centralizing or nationalizing AI would give governments fuller access to user interactions and files. To protect privacy and resilience, guests propose decentralizing inference and using confidential, verifiable stacks so agents run locally without telemetry.

“Like if it's nationalized all government, it's it's way worse than like prism and kind of because effectively this is not just metadata…”—

Agent Marketplace — Instead of a single generalist agent, they are building an agent marketplace with confidentiality and verifiability tags. Organizations can hire task specialists that run privately, access sensitive systems, and compose via blockchain coordination.

“Maybe actually going to be very expensive for it to do something.”—

Vendor Control — Model vendors have shown they can restrict capabilities on sensitive topics and change privacy policies—such as retaining paying users' data for safety. Those actions can materially reduce utility and undermine users' data expectations.

“First, Entropic released this model and kind of neutered it on on a number of dimensions and everybody was outraged about that because effectively if you asked it about cryptography…”—

MARKETS

Emerging Adoption — Stablecoins were adopted earlier in developing countries as payment rails rather than for margin. Similarly, cheaper aggregated or decentralized AI that delivers most frontier performance at far lower cost could see faster uptake in emerging markets where cost and accessibility matter more.

“Do you think one one thing that we're noticing is that stable coins got adopted in developing countries more sooner than developed countries like inside of developed countries we use…”—

Confidential TVL — Daily confidential total value locked on NEAR has been rising since the confidential TVL launch, indicating growing uptake of NEAR's confidentiality stack and private‑shard capabilities.

“This is from the Near Protocol Twitter saying daily confidential TVL on Near Intense is picking up and we're just looking at a graph going up and to the right…”—

FORECASTS

NEAR Bullish — The guest predicts that as governments make banking and transfers harder, blockchain platforms—especially privacy‑ and access‑oriented ones like NEAR—become more valuable and attractive. That underpins a bullish outlook for the protocol.

“I think on the model is near bullish the more restrictive governments become.”—

Intro

Intro frames recent controversies: Entropic restricted a model and changed privacy terms, sparking debate about who controls model use and data retention.

Pandora’s Box Opens

Pandora’s Box Opens argues open-source weights and accessible models make large-scale capabilities inevitable, enabling rapid capability arms races and vulnerability discovery.

AI Meets Government

AI Meets Government explains export controls and safety rhetoric have triggered governments to intervene, setting a precedent that could let states selectively shut down services.

Decentralizing the Stack

Decentralizing the Stack outlines technical responses: confidential inference, distributed compute, and decentralized training to resist national controls and provide redundancy.

Why Ban Fable?

Why Ban Fable? interrogates U.S. motives, suggesting weak game theory and fears of ceding tech leadership drove export controls more than a coherent safety strategy.

Internet Balkanization

Internet Balkanization shows how export controls risk fragmenting the global internet, prompting country-level internets, censorship, and pressure toward sovereign AI stacks.

KYC for AI?

KYC for AI? examines the plausible outcome where export rules force top AI labs to KYC users, creating privacy and surveillance risks and incentivizing borderless alternatives.

Nationalized Models

Nationalized Models plays forward nationalization scenarios: if governments claim models are national security assets, users may seek unregulated, KYC-free access elsewhere.

Building Private AI

Building Private AI presents Near/Ironclaw’s approach: confidential, verifiable stacks and private agents that keep user data local and resist telemetry or centralized logging.

Agent Marketplace

Agent Marketplace introduces marketplaces of specialized, verifiable agents that can be granted limited, confidential access to user data to perform tasks safely.

User-Owned Alignment

User-Owned Alignment reframes alignment as per‑user privacy and verifiability: alignment should be to an individual’s interests, not a nebulous societal standard.

Winning Against Giants

Winning Against Giants tackles product strategy: crypto projects must build UX parity (e.g., near.com) to compete with centralized incumbents while preserving privacy.

Regulation Fuels Crypto

Regulation Fuels Crypto argues restrictive policies accelerate demand for decentralized alternatives, as happened in fintech, making blockchains more attractive.

Global AI Adoption

Global AI Adoption predicts emerging markets will adopt cheaper near‑frontier models and decentralized AI faster due to cost and fewer regulatory frictions.

Compute Is the Bottleneck

Compute Is the Bottleneck identifies compute concentration at hyperscalers as the main limiter for decentralized AI, driving experiments in edge/home device aggregation.

Enterprise Use Cases

Enterprise Use Cases highlights privacy‑sensitive adopters—finance, legal, medical—and sovereign individuals and SMEs as early customers for confidential AI stacks.

Confidential TVL Explained

Confidential TVL Explained describes Near’s private shard and confidential TVL: assets held in a privacy set across multiple token types, enabling private on‑chain activity.

Full Sovereign Mode

Full Sovereign Mode outlines an end‑to‑end vision: confidential assets, private agents, and confidential inference compose a sovereign user stack for data, assets, and computation.

Related Episodes

Important Disclaimers

Not Investment Advice: For education only — not financial, tax or legal advice.

AI-Generated: Made by AI from source transcripts and may contain errors.

No Guarantees: Past performance does not predict future results; investing carries risk.

Our Commentary · Request Removal