Claude AI Discovers Critical Flaws in Post-Quantum Crypto Standard Candidate
Anthropic's unreleased Claude model uncovered a symmetry in the HAWK digital signature that slashed key recovery cost by 67 million times, potentially forcing key size doubling that undermines its viability as a NIST standard. The AI also improved an AES attack breaking a decade-old record.
Quick Take
Claude found a weakness in HAWK, reducing attack cost from 2^64 to 2^38.
Fix requires doubling HAWK keys, erasing its size advantage over other PQC candidates.
AI also sped up AES-128 7-round attack by 200-800x, breaking a 2013 record.
Findings disclosed to authors and NIST; no impact on current blockchains like Bitcoin.
Market Impact Analysis
NeutralThe finding affects a post-quantum cryptographic candidate, not current blockchain security; no immediate impact on existing crypto assets.
Speculation Analysis
Key Takeaways
- Claude Mythos Preview slashed HAWK's key recovery cost from 2^64 to 2^38 operations — a 67-million-fold reduction.
- Fixing the vulnerability forces HAWK to double its key size, erasing its compactness advantage as a NIST post-quantum candidate.
- The AI also accelerated an AES-128 7-round attack by 200-800x, breaking a 2013 record set by human cryptographers.
- Both findings were disclosed to the scheme's authors and NIST; no current blockchains are affected.
- Each experiment cost ~$100K in API usage, with Anthropic staff spending hundreds of hours verifying the AES results.
What Happened
Anthropic's unreleased Claude Mythos Preview model discovered a vulnerability in HAWK, a post-quantum digital signature scheme vying for NIST standardization. The AI identified a mathematical symmetry that human cryptographers missed, reducing the cost of stealing a private key from 2^64 to 2^38 operations. Separately, Claude improved an attack on 7-round AES-128 encryption by 200-800 times over a 2013 human record. Anthropic disclosed both findings to the scheme's authors and NIST before publishing. No deployed blockchain uses HAWK, so current crypto networks remain unaffected.
The Numbers
The attack on HAWK's smallest parameter set dropped key recovery from 2^64 to 2^38 — a 67-million-fold reduction. Fixing this forces HAWK to double its key and signature sizes, undermining its main advantage. For AES-128 7-round, the speedup ranged from 200 to 800 times the previous best attack. Each experiment cost roughly $100,000 in Claude API usage, and Anthropic employees spent several hundred hours validating the AES breakthrough.
Why It Happened
Claude exploited a symmetry in HAWK's lattice-based construction that human cryptographers had overlooked. The AI's ability to search vast mathematical spaces enabled it to find this shortcut, highlighting how ML models can augment cryptographic analysis. The AES improvement came from optimizing a divide-and-conquer technique, where Claude found a more efficient path than the 2013 manual effort. Both cases show AI's growing role in auditing cryptographic assumptions.
Broader Impact
HAWK's NIST candidacy may suffer as the fix eliminates its compactness edge over other post-quantum signatures. For blockchains eyeing quantum-resistant upgrades, larger signatures mean higher transaction costs. The finding sets a precedent: AI-assisted cryptanalysis could become standard for vetting future standards, potentially accelerating the discovery of weaknesses before deployment.
What to Watch Next
- NIST's response: whether HAWK remains in the competition after this disclosure, or if alternative schemes like CRYSTALS-Dilithium gain favor.
- Other post-quantum candidates under AI scrutiny: researchers may apply similar methods to test other lattice-based schemes.
- Impact on on-chain scaling: projects like Ethereum planning post-quantum migration will need to factor larger signatures into fee models.
This article is for informational purposes only and does not constitute financial advice.
Always late to trends?
Join for the latest news, insights & more.
Disclaimer: Bytewit is an independent media outlet that delivers news, research, and data.
© 2026 Bytewit. All Rights Reserved. This article is for informational purposes only.
