AI model exposes weakness in post-quantum crypto

TLDR

An unreleased Anthropic AI model has exposed serious weaknesses in a leading post quantum cryptography candidate, but deployed blockchain systems are not broken today.

1. Claude Mythos Preview found new attacks on the HAWK post quantum signature scheme, slashing its theoretical security and forcing much larger keys.
2. Current blockchains like Bitcoin and Ethereum still use classical signatures, so there is no immediate break, but the result shakes confidence in future quantum safe upgrades.
3. Major crypto institutions are doubling down on post quantum and AI driven security research, and developers are emphasizing crypto agility to swap algorithms when needed.

Deep Dive

1. What The AI Model Actually Broke

Anthropic reported that its Claude Mythos Preview model discovered two previously unknown attacks on post quantum algorithms, including HAWK, a NIST third round candidate for quantum resistant signatures. In HAWK, the model found a mathematical symmetry that reduced the cost of stealing the smallest key from 2^64 to 2^38 operations, making key recovery about 67 million times easier than expected. Fixing HAWK appears to require roughly doubling key sizes, undermining its selling points of compact keys and fast signing, as explained in Decrypts Claude Mythos post quantum analysis.

2. Impact On Todays Blockchains

Importantly, HAWK has never been deployed in production, and Bitcoin still uses ECDSA while Ethereum relies on ECDSA and Schnorr type signatures, so no live chain was directly broken. Crypto efforts around post quantum security, such as Vitalik Buterins roadmap, NEAR and Zcash plans, and institutional work funded by the Bitcoin Security Consortium, assumed these candidates were mathematically sound; Mythos has shown that assumption can fail late in the standardization process. Voices like Samson Mow are now warning that rushing Bitcoin into lattice based post quantum signatures could make it safe against quantum, but weaker against normal computers, urging conservative timing and favoring hash based schemes with smaller attack surfaces, as discussed in his post quantum caution.

3. How The Industry Is Responding

The finding is accelerating a broader shift where AI is used as both attacker and defender in crypto security. BlackRock, Coinbase, Fidelity and others launched the Bitcoin Security Consortium to fund open post quantum research, treating Bitcoin as critical infrastructure that needs proactive hardening. The Ethereum Foundation has elevated privacy and post quantum security in its protocol strategy and board appointments, signaling similar priorities for Ethereums long term roadmap, as noted in its SEAL 911 board move. Security leaders now stress crypto agility designing systems so signature and encryption schemes can be swapped out without rewriting everything, and watching how AI assisted attacks, not just future quantum hardware, change the risk curve.

Conclusion

Claude Mythos Preview did not break Bitcoin or Ethereum today, but it punctured a leading post quantum candidate and proved that advanced AI can find deep flaws in cutting edge cryptography. For crypto users and developers, the key takeaway is that quantum resistance is not a one time upgrade but an ongoing process, where AI assisted testing, conservative migration, and algorithm agility matter as much as the eventual choice of scheme.