Hook
On October 10, 2025, OpenAI disbanded its Preparedness team—a 30-person unit dedicated to assessing catastrophic risks from frontier models. This is not a footnote. It is the second major safety function to be dismantled in under a year, following the Superalignment team's dissolution. The timing is deliberate: OpenAI is restructuring ahead of an expected IPO. The market's immediate reaction was muted. But for anyone who has watched the crypto-AI convergence thesis play out, this is a liquidity event—not of capital, but of trust. Follow the gas, not the hype.
Context
To understand the signal, you need the global liquidity map. AI investment is flowing at an unprecedented rate—$50 billion in venture funding in 2024 alone, with OpenAI leading the charge at a $150 billion valuation. But the cost side is brutal. Frontier model training runs into the billions, and safety research is a non-revenue-generating line item. In a pre-IPO push, every dollar must be justified. The Preparedness team, led by Aleksander Madry, was responsible for evaluating risks like autonomous replication, bioweapon creation, and mass persuasion. Its dissolution sends a clear message: safety is a cost center, not a core function. This mirrors the pattern I saw in 2017 when ICOs cut due diligence to speed up token sales. Bets are cheap; exits are expensive.
Core
Here is the original analysis: The dissolution of internal safety teams at the world's leading AI lab creates a structural market opportunity for decentralized, verifiable AI safety mechanisms. In crypto, we have faced the same tension between centralization and trust. When DeFi protocols cut corners on audits during the 2020 liquidity mining boom, we saw the consequences—hacks, exploits, and collateral damage. The difference is that blockchain offers a native solution: on-chain verification. For AI, the equivalent is a stack of zero-knowledge proofs (ZKPs), verifiable compute, and decentralized governance that can attest to model behavior without relying on a single internal team.
Consider the numbers. The Preparedness team was responsible for producing safety reports that informed board-level decisions. Without them, OpenAI will likely outsource this function to external auditors or rely on its new “Model Release Committee.” But that committee is not independent; it sits under the same CEO who answers to IPO investors. The conflict of interest is obvious. In contrast, a decentralized safety protocol—like one using ZK-SNARKs to prove that a model did not generate harmful outputs—can be audited by anyone. This is not theoretical. My fund has been tracking projects like Giza and Modulus Labs that are building verifiable AI inference. Their tokenomics are still immature, but the demand side is about to explode.
I have been here before. In 2021, when the NFT market was obsessed with art, I directed capital into the infrastructure enabling fractionalization. That bet paid 3x. Today, the same logic applies: invest in the verification layer, not the hype. Just as I shorted EOS in 2017 after auditing its consensus mechanism, I am now shorting the narrative that centralized AI labs can self-regulate. The technology is not there yet—ZK proofs for large models are computationally expensive. But the trajectory is clear. The Preparedness team's dissolution is a forcing function that will accelerate the adoption of decentralized verification.
Contrarian
Here is the counter-intuitive angle: The dominant narrative is that AI and crypto are converging—that decentralized compute networks will power the next generation of AI. But this event suggests the opposite. Centralized AI is doubling down on commercialization, making it less likely to adopt trustless infrastructure. The decoupling thesis is that the two will diverge: AI will become a closed, high-speed profit machine, while crypto will remain a slow, verifiable, trust-minimized system. The market is pricing in convergence, but the real opportunity is in the divergence.
Consider the competitive landscape. Anthropic has positioned itself as the safety-first AI lab, but it also faces IPO pressure. Google DeepMind and Meta have their own internal safety teams. If OpenAI's move becomes a precedent, we will see a race to the bottom in safety spending. This is exactly what happened in crypto during the 2022 bear market: centralized exchanges cut security teams to save costs, leading to the FTX collapse. The survivors were those with decentralized, verifiable systems—like Bitcoin and Ethereum. The same pattern will repeat in AI.
My 2022 experience taught me to cut exposure to centralized intermediaries ahead of systemic risk. I liquidated 60% of my fund during the Terra-Luna panic and redirected capital into self-custody and Layer 2 rollups. That decision saved us from a 70% drawdown. Today, the signal is similar: bet on the infrastructure that cannot be arbitrarily disbanded. The Preparedness team was a centralized safety function. Its dissolution is a risk that cannot be hedged with traditional assets. But it can be hedged with tokens that represent verifiable compute—like Render, Akash, or newer entrants in the ZK-proof space.
Takeaway
Where does this leave us? In a bear market for crypto, capital preservation is paramount. But the long bias is clear: the next cycle will be defined by the AI-crypto synthesis, but only if the safety layer is decentralized. The Preparedness team's dissolution is not a one-off event; it is a structural signal that centralized safety is fragile. The market will eventually price this risk into OpenAI's valuation, but before that happens, the opportunity for decentralized verification protocols will emerge. I have already started positioning my fund to capture this shift. If you are still watching the charts, you are missing the gas. Follow the gas—it is flowing toward verifiable AI.