Claude AI exposes post-quantum crypto flaw

TLDR

Anthropics Claude AI has uncovered serious weaknesses in a leading post quantum cryptography candidate, sharpening questions about how blockchains will secure themselves against future quantum attacks.

1. Claude Mythos found attacks on the HAWK signature scheme, cutting its smallest key strength from about 2^64 to 2^38 operations and forcing much larger, less practical keys.
2. Todays Bitcoin and Ethereum signatures are unaffected, but the result shows some post quantum schemes can fail, so crypto needs multiple options and the ability to swap algorithms.
3. The key things to watch are NISTs response, Bitcoin and Zcash upgrade plans, and whether AI becomes a standard tool for stress testing crypto before it touches live networks.

Deep Dive

1. What Claude Actually Broke

Anthropics unreleased Claude Mythos Preview model discovered a structural flaw in HAWK, a post quantum digital signature scheme that NIST advanced to round three of its competition and treated as a leading candidate. In about 60 hours and roughly 100,000 dollars of compute, the model found a symmetry that reduces the work to steal HAWKs smallest key from around 2^64 to 2^38 operations, a roughly 67 million times easier attack according to a detailed write up from the competition process. That is a big drop in security for the advertised small, fast parameter set.

Fixing HAWK would require roughly doubling key sizes, which removes much of its appeal for compact signatures and low fee impact in blockchains. Anthropic disclosed the finding to HAWKs authors and it was shared via the NIST mailing list, with additional coverage from outlets like CoinDesk and Decrypt, confirming the vulnerability in this candidate scheme. Claude also produced improved attacks on a weakened AES variant and smaller, under tenfold improvements against Poseidon, a hash used in many zero knowledge systems, but those are more incremental.

Confidence: high, because the weakness is documented in public technical reporting and acknowledged by the schemes designers.

2. Impact On Cryptos Quantum Plans

Crucially for crypto holders, no major chain uses HAWK today. Bitcoin and Ethereum still rely on elliptic curve signatures, so this specific flaw does not break live wallets or transactions. The significance is that a front running post quantum contender survived years of human review and then fell quickly once a powerful AI looked at it, showing how fragile some proposed defenses can be.

Bitcoins emerging quantum plan, including proposals like BIP 360, already assumes that some algorithms will break and specifies multiple approved schemes plus fallbacks if one fails. Industry groups such as the Bitcoin Security Consortium have started funding post quantum research, treating Bitcoin security as critical infrastructure rather than a one off design. At the same time, conservative voices like Samson Mow warn against rushing into new signature schemes too early, arguing that premature migration to complex post quantum math could expose Bitcoin to easier classical attacks instead.

3. What To Watch Next

For crypto users and developers, the most important signals now are process and tooling, not emergency patching. On the process side, NISTs next communications about HAWK and its remaining post quantum candidates will shape which algorithms wallets and protocols even consider. On the tooling side, this episode is a strong proof of concept that AI models can act as cryptanalytic co researchers, meaning serious projects will increasingly use AI to attack their own schemes before shipping.

Network roadmaps that emphasize crypto agility the ability to swap signature and hash algorithms without redesigning everything from scratch are likely to gain weight. Zcashs recent Ironwood upgrade, for example, intentionally incorporates quantum recoverable notes so it can move more smoothly in a future transition. As more Bitcoin and altcoin proposals adopt that kind of flexibility, the risk of any single post quantum flaw spilling into production should fall.

Conclusion

Claudes discovery does not put current chains at immediate risk, but it is an important warning shot. Post quantum security for crypto will be shaped by an arms race where humans and AI jointly design, then attack, new schemes.

The safest path for Bitcoin and other networks is slow, conservative migration backed by diversified algorithms and strong crypto agility, with AI used as a standard tool to break candidate designs before they ever secure real value.