The Open-Source Paradox: How US AI Restrictions Echo the Mistakes of Centralized Governance

Silence is the first vote in a true consensus. But in Washington’s deliberations on open-source AI, the silence from those who build the future speaks louder than any policy memo. I watched last week as Jack Dorsey retweeted Chamath Palihapitiya’s blistering critique of proposed US restrictions on open-weight AI models. The thread went viral—not because of its celebrity, but because it laid bare a contradiction that everyone in the decentralized tech space already felt: the government wants to protect us from AI by centralizing it. Palihapitiya dropped a number that should keep every crypto founder awake at night: API costs for US firms under a closed-source regime would run $26 to $56 per million tokens, while foreign competitors running open models would pay $0.50 to $1. You don’t need a PhD in game theory to see that incentive gap. It’s a structural asymmetry that will hollow out any industry dependent on AI inference—and that means every industry. Context: The debate is framed as safety versus innovation, but it’s really about control. US policymakers, citing fears of "Mythos-level" network capabilities from models like Anthropic’s Claude Mythos, want to gatekeep the most powerful weights. Sebastian Mallaby warned that the world is about to shift from "almost nobody having dangerous capabilities to almost everybody having them." The natural bureaucratic response is to build walls. But walls in a digital world are like building a fence around a cloud. Core insight: The cost data is not the full story. Based on my experience auditing smart contracts for the DAO post-mortem in 2017, I learned that technical efficiency without ethical governance creates worse outcomes than the problems it tries to solve. The same logic applies here. A closed-source AI ecosystem would force US companies to deploy centralized models that are expensive, opaque, and ultimately less secure than open alternatives. Why? Because open-source allows for community-driven auditing—just as open ledgers allow for public verification of transactions. Block’s Goose AI agent, Dorsey’s own creation, is a perfect example of how open-source AI can be a strategic asset: its weights are public, its costs are controllable, and its alignment can be inspected by anyone. Meanwhile, a restricted API becomes a single point of failure, vulnerable to price hikes, censorship, and surveillance. Furthermore, the competitive argument is not theoretical. Beijing’s Moonshot AI released Kimi K3 this month, which took first place on a major coding benchmark. Chinese models are closing the gap not just in ability but in deployment velocity. The US restrictive approach would hand them a decades-long cost advantage. I saw the same dynamic in 2020 when designing quadratic voting for MakerDAO: the best governance system is the one that incentivizes participation, not exclusion. By making AI expensive and scarce, the US would drive its own talent and capital offshore. Contrarian angle: The opponents of open-source have a legitimate point—unrestricted release of highly capable models could enable large-scale cyberattacks, automated disinformation, and dangerous autonomous systems. But the assumption that Washington can prevent this by controlling American releases is naive. As Mallaby noted, capability diffusion is inevitable. The real choice is between a world where good actors are armed with affordable, auditable AI and a world where only bad actors have access to the cheap tools. During my six-week solitude in Hiiumaa after the FTX collapse, I realized that many of the "innovations" in crypto were just financial engineering disguised as progress. The same is happening in AI: focusing on control over deployment is a distraction from the harder work of building aligned, transparent models. A centralized AI gatekeeper would repeat the mistakes of centralized finance—opaque, fragile, and prone to regulatory capture. Takeaway: The blockchain community understands a truth that policy circles resist: security through openness is not a contradiction. Consensus requires patience, not speed. The US can either become a fortress or a garden. A fortress might feel safe today, but it starves its own growth. A garden invites participation, even from rivals, and thrives through resilience rather than restriction. As we design the governance of AGI, we must remember that code is not law—community consent is. Silence is the first vote, and it’s time we all spoke.