Anthropic’s Claude Mythos Preview model recently demonstrated a significant advancement in AI-driven cryptanalysis, successfully weakening HAWK-256, a leading candidate for post-quantum digital signatures, in roughly 60 hours. This rapid breakthrough intensifies scrutiny on the race to develop quantum-resistant cryptography, particularly for major blockchain networks like Bitcoin, which are actively preparing for a future where current encryption standards could be vulnerable to quantum attacks, highlighting the urgency for Bitcoin quantum resistance.
The discovery, announced earlier this week, highlights the accelerating pace at which artificial intelligence is influencing cryptographic security. While Bitcoin itself doesn’t use HAWK-256, the incident underscores the increasing sophistication of attacks that could impact the broader digital asset ecosystem’s transition to new, more secure algorithms.
AI’s swift cryptographic strike on HAWK-256
AI heavyweight Anthropic revealed that its Claude Mythos Preview model found a practical key-recovery attack on HAWK-256, effectively halving the scheme’s smallest key strength. The model achieved this by uncovering a previously undetected “nontrivial automorphism” within the algorithm’s lattice structure.
This flaw had eluded expert human review for two years and across two rounds of rigorous evaluation, making the AI’s rapid discovery particularly striking. The entire process, from problem initiation to discovery, took approximately 60 hours and incurred about $100,000 in computing costs, showcasing the efficiency of advanced AI in cryptanalytic tasks.
The ‘Möbius Bridge’ and other impacts
Beyond HAWK-256, Anthropic’s AI also developed a new attack technique dubbed the “Möbius Bridge.” This method made an existing attack on a deliberately weakened version of AES, a widely used encryption cipher for wallet files, between 200 and 800 times faster.
The model also yielded smaller, though still significant, improvements against Poseidon, a hash function foundational to many zero-knowledge proof systems. These systems are crucial for securing rollups and privacy protocols across various blockchain networks, signaling broader implications for cryptographic development.
Accelerating the quantum resistance race for Bitcoin
This AI-driven breakthrough aligns directly with urgent warnings from within the cryptocurrency community about the closing window for cryptographic migration. BIP-361, a companion proposal to Bitcoin’s quantum-resistant address plan, specifically argues that cryptographic attacks are improving by up to 20-fold.
Anthropic’s findings validate this trend, demonstrating that the threat isn’t solely from future quantum computers, but also from increasingly powerful classical computational methods powered by AI. Coin Center Executive Director Peter Van Valkenburgh aptly remarked on the “funniest timeline,” where Bitcoin might upgrade to a lattice-based post-quantum signature scheme only to find it already weakened by AI.
Bitcoin’s current quantum status
It’s important to note that current Bitcoin and Ethereum transactions remain unaffected by this specific attack, as both networks rely on elliptic curve signatures (ECDSA). Neither the HAWK attack nor the improved AES attack targeted these particular signature types.
However, the broader threat of quantum computing to ECDSA remains a significant concern. Research published in March 2026 by Justin Drake of the Ethereum Foundation, alongside Google Quantum AI researchers Ryan Babbush and Hartmut Neven, and Stanford’s Dan Boneh, estimated that fewer than 500,000 physical qubits could break ECDSA. This represents a 20-fold reduction from previous estimates.
Given that Google’s most advanced chip, Willow, already boasts 105 qubits, the rapid progress in quantum hardware development is clear. Estimates suggest that a quantum computer could crack Bitcoin’s cryptography in about nine minutes, a timeframe critically shorter than Bitcoin’s average block time of ten minutes.
This scenario becomes even more alarming with parallel processing, as evidenced by a 6.5 times speedup achieved with 11 primed machines running concurrently. The convergence of these threats underscores the critical need for proactive measures.
The evolving threat landscape and next steps
The incident with HAWK-256 changes the calculus for quantum migration, adding a new layer of urgency. It suggests that advanced AI could potentially undermine post-quantum encryption candidates even before fully functional quantum computers become a widespread reality, complicating an already complex transition.
The National Institute of Standards and Technology (NIST) has been at the forefront of evaluating post-quantum cryptographic schemes, including HAWK, which was advanced to the third round of NIST’s additional post-quantum signature process in May 2026. Such rigorous evaluation processes are vital in identifying vulnerabilities, regardless of whether they’re found by human experts or AI models.
This dynamic environment means cryptographic development needs to be continuously agile, anticipating threats from multiple vectors. The cryptocurrency industry can’t afford to wait for theoretical quantum threats to materialize when AI-powered classical attacks are already proving effective against prospective solutions.
The path forward for digital assets
For Bitcoin, the journey towards quantum resistance is well underway with proposals like BIP-360. This proposal specifies several NIST-standardized algorithms, offering multiple fallbacks in case one proves vulnerable to future quantum or classical advances. This multi-algorithm approach is a pragmatic response to the unpredictable nature of cryptanalytic progress.
The recent activation of privacy network Zcash’s Ironwood upgrade on Tuesday, designed to ensure recoverability of its shielded pool if quantum computers emerge, further illustrates the industry’s proactive stance. These efforts show a clear understanding that preparing for quantum threats is a multi-faceted challenge requiring continuous research and development.
As AI continues to evolve and its capabilities in code analysis and cryptanalysis grow, the collaboration between AI researchers, cryptographers, and blockchain developers will be paramount. Securing the future of digital assets against both quantum and AI-accelerated classical threats demands a dynamic, adaptable strategy.