AI has emerged as a new variable in discussions on moving Bitcoin to quantum-resistant security.

On July 28, blockchain outlet U.Today reported that Anthropic's frontier AI model discovered a new attack method against weakened cryptographic algorithms, drawing attention to whether AI could shake the mathematical foundations of post-quantum cryptography before quantum computers do.

The starting point of the debate is Anthropic's "Claude Mythos Preview" model. The model was reported to have found a new attack method against a weakened version of the Advanced Encryption Standard, or AES, a cryptographic system widely used worldwide. Anthropic said the AI-generated method increased the speed of cracking the weakened algorithm by 200 to 1,000 times.

Researchers said the result shows that frontier AI models can conduct original cryptographic research. Anthropic also said the model devised an improved attack method targeting HAWK, a post-quantum cryptographic signature scheme.

Bitcoin is drawing attention again at this point. Bitcoin currently uses ECDSA for transaction approvals, and the cryptography community has long warned that ECDSA could eventually be broken if sufficiently powerful quantum computers emerge. Bitcoin developers have therefore been reviewing ways to move to post-quantum signature schemes, and lattice-based cryptography has been cited as one of the leading candidates.

The research does not mean an immediate threat to Bitcoin or internet security more broadly. Researchers stressed that the target was not a cryptographic system in a real operating environment but a scheme intentionally weakened for research. Anthropic also said modern AES remains safe.

Market and industry attention is focused on how quickly AI can push cryptographic research forward. Peter Van Valkenburgh (피터 반 발켄버그), executive director of Coin Center, said a situation could arise in which, after Bitcoin upgrades to a lattice-based post-quantum signature scheme, advanced large language models break the structure before quantum computers become a threat. His remark reflects the view that, when designing security models, quantum threats cannot be considered in isolation, rather than implying a risk that has already materialised.

The key issue is that Bitcoin's security transition may not be limited to responding to quantum computers. If AI systems continue to accelerate mathematical discoveries and cryptanalysis, developers of Bitcoin and other blockchain networks may need to assess AI-based cryptanalysis alongside quantum decryption capabilities when reviewing post-quantum security.

Against that backdrop, discussions over Bitcoin's next-generation signature scheme are expected to become more complex. That is because it will be necessary to weigh both when quantum computers could threaten current Bitcoin signatures and how quickly AI can find weaknesses in future cryptographic systems. The network is not in danger right now, but the chance that the criteria for a post-quantum shift will change has increased further.

Funniest timeline: Bitcoin upgrades to a post-quantum lattice signature scheme... ...and then frontier LLMs break the lattice math before quantum computers ever break ECDSA. Anthropic shows that AI advancing mathematics could equal the threat of quantum advancing computation…