An artificial intelligence system has weakened a proposed post-quantum signature algorithm in about 60 hours, intensifying scrutiny of the cryptographic tools being considered to protect Bitcoin and other digital systems from future quantum computers.Anthropic’s Claude Mythos Preview identified a mathematical weakness in HAWK, an experimental digital-signature scheme competing in a US government process to select additional post-quantum security standards. The discovery does not compromise Bitcoin, which does not use HAWK, and it affects no deployed software or financial network.
The finding nevertheless carries broader implications for Bitcoin developers, who are examining how the network should replace or restrict the elliptic-curve signatures that could eventually be defeated by sufficiently powerful quantum machines.
HAWK had passed two rounds of review by specialist cryptographers over roughly two years. The AI-assisted work found an overlooked symmetry in the mathematical lattice underpinning the system, allowing researchers to recover keys substantially faster than previously calculated.
The expected computational cost of attacking the smaller HAWK-256 parameter set fell from about 2 to the power of 64 operations to roughly 2 to the power of 38. Larger HAWK configurations remain beyond practical attack, while the technique does not undermine lattice cryptography generally or other post-quantum candidates.
Researchers said restoring HAWK’s intended security level would require significantly larger keys, weakening the efficiency advantages that helped make it attractive. The scheme’s designers were informed before public disclosure, enabling the findings to be assessed through the normal review process for cryptographic standards.
Claude worked through published research, mathematical reasoning and computational tests in a multi-agent environment, with occasional guidance from a researcher. Discovering, implementing and verifying the HAWK attack took about 60 hours and generated an estimated $100,000 in computing costs.
The same AI system also improved an attack on a deliberately weakened, seven-round version of AES-128, cutting the workload by an estimated factor of 200 to 800. Standard AES-128 uses 10 rounds and remains unaffected, meaning the result is primarily significant as evidence that frontier AI models can perform advanced cryptanalysis.
For Bitcoin, the development highlights a difficult assumption behind any migration to post-quantum security: replacement algorithms must withstand sustained examination by human researchers, AI systems and potentially new mathematical techniques emerging before quantum hardware becomes operational.
Bitcoin currently relies on ECDSA and Schnorr digital signatures to prove ownership and authorise spending. These systems remain secure against conventional computers but could be broken using Shor’s algorithm on a cryptographically capable fault-tolerant quantum machine.
The greater immediate concern involves coins whose public keys are already visible on the blockchain. Bitcoin addresses can conceal a public key until funds are spent, but older payment formats, reused addresses and some modern output types expose keys that a future quantum attacker could use to calculate corresponding private keys.
A draft proposal known as Bitcoin Improvement Proposal 361 seeks a phased migration away from vulnerable signatures after a post-quantum transaction format has been selected. Its first stage would prevent users from sending coins to quantum-vulnerable addresses, while a later stage would impose tougher conditions on spending coins secured by ECDSA or Schnorr.
The proposal envisages roughly three years between activation and the first restriction, followed by another two years before the second stage. It remains a draft and depends on a separate proposal defining an accepted post-quantum signature mechanism.
That approach has divided developers and investors because dormant wallets may contain coins whose owners have died, lost their keys or cannot be contacted. Restricting vulnerable outputs could leave substantial holdings permanently inaccessible, including coins widely believed to have been mined by Bitcoin’s pseudonymous creator, Satoshi Nakamoto.
Allowing those coins to remain spendable, however, could enable the first actor possessing a capable quantum computer to seize them. Developers must therefore weigh property rights, network stability and technical security while reaching agreement across miners, exchanges, wallet providers, custodians and individual users.
Post-quantum signatures also tend to be much larger than Bitcoin’s existing signatures. That could increase transaction fees, reduce effective block capacity and add storage and bandwidth requirements for nodes, raising the prospect that quantum protection could revive disputes over network scale and operating costs.
Quantum hardware is not presently capable of breaking Bitcoin signatures, and estimates for when such machines may emerge vary widely. The network’s challenge is that designing, testing and deploying a replacement may take years, while millions of users and service providers would need sufficient time to move their holdings.
Topics
Cryptocurrency