Imagine a world where the code that powers the internet’s most transformative technology is locked behind proprietary gates, accessible only to those who can afford the highest toll. That’s the debate raging in AI right now—a battle between those who see open-source models as a security nightmare and those who see them as the only path to sustainable innovation. This isn’t a new fight. We’ve had it in crypto for over a decade, and the outcomes hold harsh lessons for the AI industry.
The Debate That Refuses to Die
Last week, a conversation between Jack Dorsey, Chamath Palihapitiya, and David Sacks resurfaced a core question: Should the US restrict the release of powerful open-source AI models? Policymakers in Washington argue that models like Anthropic’s Claude Mythos—a system that has raised “Mythos-level” cyber capabilities concerns—pose an existential national security risk. They want to limit what can be made public, fearing that latent capabilities, once widely available, will be exploited by state-sponsored hackers and malicious actors.
But the open-source camp fights back with stark economics. Palihapitiya dropped a specific claim: “If we close the open-source AI ecosystem, US companies will pay $26 to $56 per million tokens for API access, while their overseas competitors will pay $0.50 to $1.” That’s a 26-to-56-fold cost disadvantage. He called it “unsustainable,” especially if AI is truly the engine of future economic activity. David Sacks, a lifelong tech investor and now a policy advisor, added that the solution isn’t restriction but accelerated AI-driven defense—a classic “fight fire with better fire” approach.
I’ve seen this script before. In 2017, I was auditing early Ethereum whitepapers for a consultancy called EthicalChain. We identified Ponzi schemes disguised as decentralized exchanges, and our findings went viral through Telegram groups. That experience taught me that code isn’t law—it’s a social contract. And when you try to lock down that contract, you create perverse incentives.
The Cost Deception
Let’s unpack that cost claim. $56 per million tokens sounds outrageous compared to $0.50. But hidden in that gap is a deeper story: closed APIs bundle convenience and capability, but they also bundle power. When you use a proprietary model, you’re not just paying for inference—you’re paying for a walled garden. The overseas competitors Palihapitiya references are likely using open-weight models like Meta’s Llama 4 or China’s Kimi K3 (which just topped a prominent coding benchmark) running on their own low-cost hardware, possibly subsidized by national AI infrastructure.
Based on my years watching Layer-2 scaling debates, I can tell you the same dynamic applies. Post-Dencun, blob data will be saturated within two years, and rollup gas fees will double. Centralized sequencers might offer lower costs now, but they sacrifice trust. Here, the open-source AI model is like an L2 that forces users to run their own full node—higher upfront cost, lower long-term risk of censorship and price gouging.
But the cost isn’t just monetary. Palihapitiya also pointed out a security asymmetry: the US pays $56 per million tokens to defend its critical infrastructure, while adversaries pay $0.50 to attack. That’s a 70-to-1 cost ratio—defenders are priced out of their own safety. In crypto, we see the same imbalance: a small exchange needs millions to build security, and a single hacker can drain it for pennies on the dollar.
When Code Becomes Conscience
I remember curating SoulBound Stories in 2021, a digital art exhibition where NFTs could not be sold—only gifted. That project taught me that technology doesn’t just transfer value: it embodies identity. An open-source AI model, once released, becomes a shared artifact. It can be fine-tuned by a researcher to detect cancer, or by a regime to automate surveillance. The code itself is neutral, but its distribution is a moral choice.
This is the heart of the debate. The open-source camp, led by Dorsey and Sacks, argues that widespread access is a net good because it democratizes power and accelerates collective defense. The restriction camp fears that “Mythos-level” capabilities—things like autonomous tool creation, stealth network breaches, or self-improving architectures—should not be placed in every developer’s hands. They believe in a guardian class of responsible AI labs.
But history suggests guardians often become gatekeepers. In 2017, I audited a project whose whitepaper claimed to be “fully decentralized” but had a single admin key on a multi-sig wallet controlled by three co-founders. That’s not code is law; that’s code is permission. Similarly, if the US restricts open-source AI, the models will still be developed—by China, by rogue labs, by anyone with a GPU farm. Sebastian Mallaby captured this: “The world will soon go from almost no one having this power to almost everyone having it, regardless of policy.”
A Blockchain Blueprint for AI Governance
So where does that leave us? I believe the answer lies not in choosing open or closed, but in building a new governance layer—something blockchain natives call a “sociotechnical stack.”
During my 2022 bear market pivot, I launched a 10-part series called “Surviving the Winter” that reached 50,000 readers. That series emphasized that resilience isn’t about avoiding losses but maintaining faith in the decentralized ethos. The same applies to AI. We need an ethos that says: trust the verification, not the gatekeeper.
What if every major AI release came with a cryptographically signed provenance record—verifying the model weights, the training data footprint, and the safety evaluation results? What if we used zero-knowledge proofs to let developers use a model’s capability without revealing the full weight set? That would preserve the openness we need for competition while reducing the risk of one-click weaponization.
This isn’t just fantasy. My current project, TruthLayer, timestamps AI-generated content on a blockchain to combat deepfakes. We’ve secured seed funding because businesses want verifiable truth. The same concept can scale: imagine a decentralized registry of AI model disclosures, where any breach of safety commitments is transparent and punishable through smart contracts.
The Contrarian: Is Open Source a Mythos Risk Too Far?
I need to be honest. I’ve been an open-source evangelist for years—“Democracy isn’t a transaction where every voice holds weight,” I often write. But the AI danger is different from crypto. A buggy DeFi contract loses money. A buggy AI model with autonomous capabilities could cause harm on a different scale—power grid sabotage, mass disinformation, autonomous weapons.
David Sacks’ proposed solution—AI-driven defense—presupposes that the good side can innovate faster than the bad side. That’s the same assumption that drove the crypto arms race between Layer-1 security and exploit developers. Sometimes the attackers win. The DAO hack wasn’t stopped by code; it was stopped by a contentious Ethereum fork. We don’t have a fork button for AI models once they’re trained and released.
Perhaps a middle path exists: release only “filtered” open models that have been safety-tuned to remove the most dangerous capabilities, while keeping the core utility. But as Palihapitiya warned, such restrictions would still create a two-tier system where US firms pay a premium for the filtered version while overseas pirates run the unfiltered original at near-zero cost.
Takeaway: The Fork in the Road
Jack Dorsey, Chamath Palihapitiya, and David Sacks are not merely debating AI policy—they are reliving the foundational debate of crypto: permissioned vs permissionless. In 2009, Bitcoin chose permissionless. It hasn’t always been pretty, but it has been resilient. Today, AI faces a similar choice. Will we build a future where power is distributed, even if that means occasional misuse? Or will we build one where a few labs, backed by sovereign states, hold the keys to the most profound technology since fire?
I don’t have a perfect answer. But I know that “Scarcity creates meaning. Supply creates noise.” Restrict AI too much, and you create an artificial scarcity of intelligence that enriches a few and weakens everyone else. Release it without guardrails, and you risk noise becoming cacophony. The blockchain path offers a third way: transparency, verifiability, and community-rooted governance. It’s not a panacea—but it’s the only path that respects both freedom and responsibility.