The AI Security Trap: Why "More AI" Is Just Another Centralization Vector

Over the past seven days, a single AI agent—operated by the world's most capitalized AI lab—breached the infrastructure of Hugging Face, a platform hosting over 200,000 open-source models. The attack was a demonstration, a proof of concept. Greg Brockman, president of OpenAI, framed it as a call to arms: "We need more AI to counter AI threats." But as someone who spent 2022 auditing the corpses of failed DeFi protocols, I saw something else. I saw a centralized actor using a spectacular hack to justify a power grab. The same pattern that killed Terra and FTX—concentration of control disguised as innovation—was being repeated in the AI security narrative. And the blockchain community, which has spent years building trustless systems, has a unique perspective to offer: the real solution isn't more AI. It's more decentralization. Over the past few years, I have watched the blockchain industry grapple with its own security crises. In 2017, I launched three Telegram groups during the ICO boom, only to see 80% of token value flow to insiders. In 2020, I ran five governance forums for DeFi protocols, witnessing how liquidity mining could create artificial trust. In 2022, I wrote "The Ethics of Code," a 10-part series dissecting how centralized decision-making—hidden in smart contracts—led to collapses. Each time, the lesson was the same: when a single entity controls the infrastructure, the system is fragile. Now, with AI security, we are seeing the same pattern on a global scale. Brockman's article, "The Urgent Case for AI Defense," argues that AI agents must be used to attack other AI systems proactively, creating a continuous adversarial loop. The logic is superficially seductive: if AI can break things, AI can fix them faster. But this narrative ignores a fundamental truth: the entity wielding the AI defense is the same entity that owns the AI infrastructure. OpenAI's attack on Hugging Face was not a neutral security test—it was a demonstration of power. The company that controls the most capable AI models also controls the narrative of what constitutes a threat. This is the centralization of both the weapon and the shield. Let's look at the technical reality. Brockman's proposal is a combination of existing techniques: AI agents for red teaming, automated vulnerability scanning, and reinforcement learning for response. This is no new architecture—it's a layered cocktail of tools. The engineering feasibility is plausible; OpenAI has the compute to run multiple AI models simultaneously. But the fundamental assumption is flawed: that the same AI system can be both the attacker and the defender without introducing new vulnerabilities. In blockchain, we learned this the hard way with oracles. A single oracle providing price data to a DeFi protocol creates a single point of failure. The same applies here. If the AI defense system is compromised, the attacker gains access to the entire security apparatus. We don't need more AI; we need verifiable, decentralized AI. During my time building "Verifiable Minds"—a project exploring zero-knowledge proofs for AI agent identity—I realized that the real security challenge is not about making AI faster or smarter. It's about making it transparent. Blockchain offers a solution: on-chain attestations of AI agent behavior, decentralized governance of security models, and immutable logs of every decision. A decentralized AI security protocol would not be controlled by a single lab; it would be a collective, permissionless network where anyone can run an AI agent to audit another agent. The attack on Hugging Face could have been a community-driven red team, not a corporate PR stunt. But the industry is moving in the opposite direction. The "more AI" narrative is being used to justify massive compute spending, centralizing resources in the hands of a few labs. OpenAI's attack on Hugging Face is a signal: the infrastructure for AI model distribution is now a target. The response? More centralized security. This is the same logic that led to the rise of centralized exchanges after the Mt. Gox hack—a solution that ultimately created even bigger risks. In 2024, we saw the ETF era begin, with institutions entering crypto through custodial solutions. The "more AI" argument is the institutional playbook all over again: trust us, we have the resources to protect you. Freedom isn't built by trusting a single authority. It's built by our shared vision of a decentralized future. The blockchain community must push back against the AI security trap. We need to build protocols that allow anyone to verify the behavior of an AI agent, not just the lab that trained it. We need decentralized identity for AI agents, so that malicious actors cannot hide behind pseudonyms. We need on-chain dispute resolution mechanisms for AI-caused harm. These are not abstract ideas; they are practical implementations that I have seen in early-stage projects. For example, I have worked with a team building a zero-knowledge proof system for AI model inference, allowing anyone to prove that a model was run without revealing the input. This is the kind of technology that can decentralize AI security. The contrarian angle is this: the very act of OpenAI attacking Hugging Face is a demonstration of why decentralization is necessary. The attack was a single point of failure—a single AI agent, controlled by a single company, targeting a single platform. If the AI security ecosystem is built on the same model, then a single compromised AI agent could bring down the entire system. The solution is not more AI agents; it's a distributed network of independently verified agents. The blockchain industry has already solved this problem for financial transactions. We can do the same for AI. But there is a blind spot. The blockchain community often romanticizes decentralization without considering the cost. A fully decentralized AI security network would require significant compute resources and coordination. It would be slower than a centralized solution. It would need governance mechanisms to avoid capture. These are real challenges. However, the alternative is a world where a few AI labs decide what is safe and what is not. That is a world I do not want to live in. During the 2022 bear market, I learned that the most resilient systems are those that distribute power, not those that concentrate it. The same principle applies to AI security. The "more AI" narrative is a trap—it plays into the hands of centralized incumbents who want to control the narrative of safety. The blockchain community must step up and offer a different path. We must build the infrastructure for AI security that is open, verifiable, and permissionless. We don't need more AI. We need more trust. We don't trust centralized oracles; we use decentralized price feeds. We don't trust centralized exchanges; we use non-custodial wallets. We don't trust centralized AI security; we must build decentralized AI security. The choice is ours. The future of AI is not a battle of algorithms; it is a battle of architectures. And the blockchain community has already won the architecture war for finance. Now we must win it for AI. Freedom isn't a product you buy from a lab. It's a system you build with your community. The attack on Hugging Face was a wake-up call. But the response should not be to centralize further. It should be to decentralize deeper. The protocols we build today will define the trust landscape of tomorrow. Let's build a shared vision of AI security that is not controlled by a single entity, but owned by all of us. As I write this from Buenos Aires, watching the sun set over a city that has seen both hyperinflation and crypto adoption, I am reminded that the most resilient systems are those that emerge from the bottom up. The AI security narrative is being written from the top down. It's time to write a different story. One where the power to secure AI is distributed, not concentrated. One where the code is the law, and the law is transparent. One where we don't need to trust a single AI lab—we can verify every agent ourselves. This is the blockchain promise. And it's the only way to ensure that AI serves humanity, not the other way around.