The AI Crackdown Could Change the Internet Forever

Key Concepts

• User-Owned AI: A paradigm where AI models, data, and agents are controlled by the user rather than centralized corporations or governments, ensuring privacy and alignment with individual interests.
• Confidential Inference: A technical framework allowing AI models to process data without exposing the underlying information to the service provider or third parties.
• Export Controls: Government-imposed restrictions on the distribution of advanced technology (e.g., Fable 5), which the speaker argues sets a dangerous precedent for internet censorship.
• Sovereign AI: The concept of nations or individuals maintaining control over their own AI infrastructure to avoid reliance on foreign-controlled centralized services.
• Confidential TVL (Total Value Locked): A metric tracking assets held in private, encrypted shards on the NEAR blockchain, ensuring transaction privacy.
• Agent Marketplace: A decentralized ecosystem where specialized AI agents can be hired to perform tasks using private data without risking data leakage.

---

1. The Impact of Government Export Controls

The discussion centers on the recent US government export ban on the "Fable 5" AI model, citing safety and security concerns. The speaker, Ilia, argues that this action is a "strange precedent" that effectively treats AI services like a switchable utility, similar to how some nations have blocked internet access.

• Key Argument: This move risks "balkanizing" the internet, forcing other nations to develop their own sovereign AI labs to avoid dependency on US-controlled technology.
• Economic Perspective: The speaker suggests that such restrictions will likely accelerate the development of decentralized AI alternatives, as nations and developers seek to bypass centralized bottlenecks.

2. The "Pandora’s Box" of AI Capabilities

Ilia asserts that the era of high-intelligence AI is inevitable and already underway.

• Open Source vs. Closed: While frontier models like Fable 5 are powerful, open-source and open-weight models are rapidly closing the performance gap.
• Security Reality: The speaker notes that malicious actors can already use smaller models (e.g., 7B parameter models) to find vulnerabilities. He argues that policing "thoughts" (what people ask AI) is ineffective; instead, society should focus on securing physical production chains and using AI to defend against AI-driven threats.

3. The Case for User-Owned and Aligned AI

A central theme is the misalignment between centralized AI labs (like Anthropic or OpenAI) and the end-user.

• Economic Incentives: Ilia explains that centralized labs are driven by revenue optimization (e.g., AB testing for ad revenue), which often conflicts with providing the best user experience.
• The "Mom" Analogy: He defines ideal AI alignment as a system that is always on the user's side—acting as a supportive partner that provides critical feedback when the user is about to make a mistake, rather than a "sycophantic" bot that simply agrees with everything.
• Privacy Risks: Centralized agents often have access to sensitive data (passwords, healthcare records, financial transactions). If these labs are nationalized, the government could theoretically access every private interaction, creating a "1984-style" surveillance state.

4. Methodologies for Decentralized AI

To counter centralization, the speaker outlines a framework for "User-Owned AI":

• Confidential Inference: By running models in a verifiable, encrypted environment, users can provide context (meeting notes, emails) to agents without the agent provider ever seeing the raw data.
• Agent Marketplace: A decentralized platform where users can hire specialized agents. Because these agents run on a verifiable, confidential stack, they can be granted access to private CRM or financial data without the risk of data leakage or unauthorized telemetry.
• Load Balancing Compute: Decentralized networks can optimize compute usage by routing tasks to regions where compute is currently underutilized (e.g., tapping into global slack in the system).

5. Synthesis and Conclusion

The conversation concludes that the current trend of government restriction and corporate centralization is the primary catalyst for the growth of decentralized technologies.

• Actionable Insight: The speaker views the "sovereign individual" (prosumers, small-to-medium enterprises) as the primary target audience for these tools.
• Final Takeaway: As governments become more restrictive, the value proposition of decentralized, user-owned AI increases. By building a "vertical stack" from blockchain-based asset management to confidential AI inference, the NEAR ecosystem aims to provide a "full sovereign mode" where users retain total control over their data, assets, and AI interactions, effectively creating a resilient, borderless alternative to the centralized internet.