AI model exposes flaw in post-quantum crypto

TLDR

Anthropics Claude Mythos AI model has exposed serious weaknesses in a leading post?quantum signature scheme, raising new questions about how crypto will defend against both quantum and AI attacks.

1. Claude Mythos found an attack that roughly halves the security of the HAWK post?quantum signature scheme, leading its authors to withdraw it from a US standards process.
2. No blockchain or banking system broke, but the result shows that some quantum?safe designs can fail under classical and AI?assisted cryptanalysis, complicating Bitcoins and others migration plans.
3. Crypto users should watch NISTs post?quantum standards, protocol upgrade debates, and the rise of crypto?agility so networks can swap algorithms quickly if AI finds more flaws.

Deep Dive

1. What Was Broken

Anthropic reports that its unreleased Claude Mythos Preview model discovered new attacks on several cryptographic algorithms, including HAWK, a lattice?based post?quantum digital signature candidate in NISTs competition. Mythos found a mathematical symmetry that cuts the work to recover HAWKs smallest key from about 2^64 to 2^38 operations, effectively halving its security level and removing much of its efficiency advantage for blockchain use. This is documented in Anthropics write?ups and coverage such as Claude Mythos cracked post?quantum cryptography and Anthropic says Claude found new weaknesses in cryptographic algorithms.

Following coordinated disclosure, HAWKs designers notified NIST that they were withdrawing the scheme from the additional signature process, as summarized in a standards?focused community report. Claude also sped up a known attack on a deliberately weakened seven?round version of AES?128 by 200 to 800 times, but full ten?round AES used on todays internet remains unbroken, as noted by both Anthropic and the New York Times.

2. Impact On Bitcoin And Crypto

Importantly, nothing currently deployed on major chains was directly broken. Bitcoin and Ethereum still use ECDSA and Schnorr signatures, and Anthropic stresses that SHA?256, ECDSA and standard AES are unaffected in production systems, as summarized in this technical overview.

The deeper impact is on the roadmap. Bitcoin developers have been drafting plans to move to post?quantum signatures, assuming NIST?chosen schemes would be safe enough, and BIP proposals explicitly mention the need for fallback options if an algorithm later fails. Mythoss result shows that even after years of human review, a frontier AI model can find weaknesses fast and cheaply. That has sparked debate among cryptographers and Bitcoin builders about not rushing into lattice?based schemes and instead prioritizing crypto?agility systems that can swap algorithms without redesigning whole networks, a position echoed by commentators like Samson Mow and Ledgers CTO in recent discussion of AI?found flaws.

3. What To Watch Next

Several threads are worth watching. First, NISTs remaining post?quantum signature candidates and how often they attract intense AI?assisted scrutiny, given HAWKs withdrawal after Mythoss attack, as noted in standards coverage. Second, how Bitcoin and other large networks encode crypto?agility into wallets and protocols so they can roll out new signature schemes without mass breakage.

Third, regulators and financial institutions are starting to measure quantum and post?quantum readiness, including blockchain settlement, through initiatives like Hong Kongs Quantum Preparedness Index and central bank experiments with post?quantum signatures, which highlight that tokenized assets and crypto rails are part of the same long?term risk surface as banking encryption.

Conclusion

Claude Mythos did not break Bitcoin or current internet encryption, but it did knock out a prominent quantum?safe candidate and accelerate attacks on weakened ciphers, proving that AI is now a serious cryptanalysis tool. For crypto, the key takeaway is that post?quantum migration is not just about picking one new algorithm; it is about building flexible systems that can survive both quantum progress and AI?driven discoveries of flaws.