Common Knowledge Base Association (CKBA) (@ckba_build): Anthropic just announced that its Claude Mythos Preview model discovered a new attack that significantly weakens HAWK, a candidate post-quantum signat…

Anthropic just announced that its Claude Mythos Preview model discovered a new attack that significantly weakens HAWK, a candidate post-quantum signature scheme. HAWK had survived two rounds of expert review over two years. Mythos then spent roughly 60 hours synthesizing known cryptanalytic techniques and found an attack that effectively halved its key strength. To be clear, the attack does not make HAWK practically breakable today, and the scheme has not yet been deployed. But restoring its intended security would require roughly doubling its key sizes, undermining the efficiency that made it attractive in the first place. But there is a much bigger lesson here than an AI model weakening one algorithm. In crypto, the post-quantum transition is often framed as a one-time upgrade: replace ECDSA with a quantum-resistant signature scheme, and the problem is solved. The reality—and what @AnthropicAI’s result reinforces—is that there is no perfect post-quantum signature scheme waiting for us. Every option involves trade-offs, different security assumptions, and the possibility that future cryptanalysis will expose weaknesses we do not yet understand. And this was not a quantum attack. It was AI-assisted cryptanalysis. Just imagine what could happen five, ten, or fifteen years from now, when cryptographically relevant quantum computers are combined with the AI models of the future. We do not intend to fear monger, but it’s quickly becoming super clear that the solution isn’t a one-time upgrade. Instead, as the good folks at @projecteleven point out, the key is crypto agility: the ability to adopt new cryptography, support multiple schemes simultaneously, and migrate away from weakened algorithms without redesigning the entire network. So far, CKB is the only blockchain that can do that. On CKB, signature schemes are implemented in programmable Lock Scripts rather than hardcoded as consensus-level primitives. New schemes can be deployed as code, multiple schemes can coexist, and users can migrate at their own pace without waiting for a network-wide fork. CKB already supports SPHINCS+ on mainnet, but if Anthropic—or anyone else—finds a serious weakness in it tomorrow, CKB is ready to replace it with something better like 🫰that.