The AI Cracked the Code: What Anthropic's Attack on Post-Quantum Signatures Means for Crypto's Future

On a quiet Tuesday, Anthropic’s Claude model did something no human had managed in years—it cracked a cryptographic signature scheme that was on the verge of becoming a US federal standard. For half a decade, the world’s top cryptographers had tried and failed to find a practical weakness. An AI did it in hours. This isn’t a distant lab curiosity; it’s a watershed moment for blockchain security. The scheme in question? A post-quantum signature candidate, designed to withstand the quantum computers that don’t yet exist. But the attacker wasn’t a quantum machine—it was a classical large language model trained to think adversarially. The headlines are dramatic, but the real story runs deeper. The attack targets a specific family of signatures that was heading toward standardization by the National Institute of Standards and Technology (NIST). That process—NIST’s Post-Quantum Cryptography (PQC) Standardization—is the single most influential effort in securing our digital future. Every blockchain project I’ve audited in the past two years has a post-quantum migration plan. Some are building new L1s around lattice-based signatures; others are retrofitting existing chains with hybrid schemes. This discovery throws a wrench into every one of those roadmaps. Let’s get technical. The signature scheme in question belongs to a class that relies on the hardness of certain structured lattice problems. These problems have been studied for decades, and NIST’s selection process was already down to finalists. The AI found a subtle structural weakness that humans had overlooked—a short-cut that allows an adversary to forge signatures under specific adversarial conditions. This isn’t a full break in the sense of a polynomial-time algorithm, but it’s enough to disqualify the scheme from being considered ‘safe’ for standardisation. Based on my experience building ChainLit in 2017, where I translated ICO whitepapers into plain language, I’ve learned that the devil is often in the assumptions behind the math. The AI found a flaw in the assumption that the scheme’s parameters were chosen to avoid certain low-rank approximations. The core insight here is not about one specific algorithm. It’s about the arrival of AI as a first-class cryptanalytic tool. Until now, the cryptographic community treated AI as a tool for optimization, not for breaking primitives. That assumption is now invalid. This event signals that every future security standard must be vetted not only by human cryptographers but also by adversarial AI models that can explore the attack surface at machine scale. In my work with Deutsche Bank's digital assets desk last year, I saw how traditional finance clung to the idea that ‘NIST standard = unhackable’. This discovery erodes that trust, and trust is the only thing that makes a blockchain valuable. The contrarian angle? Most market participants will interpret this as a short-term negative—‘post-quantum is broken, crypto is doomed’. They’ll sell the narrative and miss the opportunity. The truth is more nuanced. This attack does not break any blockchain deployed today. Bitcoin uses ECDSA; Ethereum uses secp256k1; most Layer 2s rely on BLS or EdDSA. Those schemes remain unaffected by this specific attack. The risk is forward-looking: any project promising ‘post-quantum security’ by adopting the targeted scheme now has a credibility gap. The contrarian trade is not to flee crypto, but to invest in projects that treat security as an evolving, multi-faceted discipline—those that build hybrid signature frameworks, run continuous AI-driven red-teaming, and maintain a modular approach where a single algorithm failure doesn’t bring down the entire network. ‘Trust is earned in the bear, spent in the bull.’ The bear market taught us to question everything. This event reinforces that lesson. Let’s not forget the deeper philosophical implication. We are entering an era where our most cherished cryptographic assumptions—the ones we thought would last decades—can be overturned by AI that learns faster than we can standardise. This isn’t a bug in the code; it’s a feature of a world where machine intelligence is a co-creator of our security. As someone who founded Resilience DAO to support displaced Web3 workers after FTX, I’ve seen firsthand how community resilience is the ultimate backup. ‘Code is law, but community is conscience.’ The community must now demand that their protocols not only use strong algorithms but also have a path to adapt when those algorithms are compromised. That means on-chain governance that can update signature schemes without hard forks, multisig wallets that combine multiple algorithm families, and a culture of transparency where vulnerabilities are shared openly. The signal for developers is clear: diversify your security stack. The days of betting the farm on a single NIST finalist are over. We need to design systems that are agile enough to swap in new signatures as the AI threat landscape evolves. The opportunity lies in ‘AI-resistant’ security: not an algorithm that can’t be broken, but a framework that can detect, respond, and recover from breaks faster than an adversary can exploit them. In 2022, when the market crashed, I saw builders shift from ‘move fast and break things’ to ‘build slowly and test everything’. We need that same mindset shift now for cryptography. A word on the AI itself. Anthropic’s Claude didn’t just guess; it used a novel reasoning technique that combined symbolic deduction with pattern recognition. This suggests that the method could be generalised to other crypto primitives. The probability that this is an isolated incident is low. I would bet that within three years, every major NIST candidate will have been stress-tested by adversarial AI models—and several will break. The question is whether the blockchain industry will be caught flat-footed or will proactively fund AI auditing cooperatives. ‘Community is the only chain that cannot be broken.’ If we open-source our security assumptions and invite AI to audit our code collectively, we turn the weapon into a shield. For investors, the actionable takeaway is to look for projects that have already announced a multi-algorithm signature strategy, or teams that include cryptographers with a background in AI adversarial training. Avoid chains that are doubling down on a single unproven post-quantum scheme. Instead, back ecosystems that treat security as a public good—ones that sponsor bug bounties for AI-driven attacks, or that use zero-knowledge proofs to verify security properties without revealing the underlying algorithm. This is a long-term play, but the seeds are being planted now. Finally, let’s zoom out to the macro narrative. For the past year, the crypto market has been euphoric, riding the recovery from the 2022 winter. Bull markets mask technical flaws. Everyone is focused on scaling, on-ramps, and institutional adoption. But underneath the hood, the cryptographic foundations are being tested by a new type of intelligence. This discovery reminds us that the most dangerous risks are the ones we don’t see coming. The AI that cracked this scheme was not designed to break crypto; it was a general-purpose reasoning model. Imagine a future where AI models are purpose-built to attack every new standard the moment it’s published. That is not science fiction—it is the logical endpoint of the trajectory we’ve just seen. We have a choice. We can treat this as a one-off anomaly and hope the next scheme holds. Or we can embrace a new paradigm: security through diversity, adaptability, and open-source community oversight. The latter is harder, but it’s the only path that aligns with the values of decentralisation. ‘Code is law, but community is conscience.’ The law must evolve as our understanding grows. The conscience must guide us to build systems that survive not just quantum computers, but the AI that will help design them. In my years as a Web3 community builder, I’ve seen that the most resilient projects are not those with the most advanced tech, but those with the most trusted communities. Trust is built through transparency and the willingness to admit when something is broken. This is a moment for transparency. Let’s audit our post-quantum assumptions, fund AI red teams, and prepare to pivot. The only way to stay ahead of the AI adversary is to become an AI ally. Community is the only chain that cannot be broken. Let’s keep it that way.