HAWK is not deployed anywhere, nor is it used by Bitcoin. Still, it sits within the wider race to develop quantum-resistant cryptography—a challenge Bitcoin developers are working urgently to address.

This week, AI firm Anthropic revealed that its Claude Mythos Preview model found a weakness that effectively halves HAWK’s key strength. HAWK is a proposed successor to the digital signature systems that secure online banking and web transactions.

The result came after roughly 60 hours of work and about $100,000 in computing costs, breaking into an algorithm that had already undergone two years of development and multiple rounds of expert review.

Digital signature schemes like HAWK confirm that data or messages come from the holder of a specific private key. Bitcoin uses such signatures for every transaction, and they also underpin secure browsing, as indicated by the padlock icon in web browsers.

Today’s signature systems are expected to fail against future quantum computers, and HAWK is one of several candidates under evaluation. However, it has not yet been deployed in real-world systems.

The findings do not impact Bitcoin or Ethereum at present, as both rely on elliptic curve cryptography, which was not targeted. HAWK is also not among the schemes Bitcoin intends to adopt.

Bitcoin’s BIP-360 proposal outlines a transition to quantum-resistant addresses using three NIST-standardized algorithms, with multiple options included to provide fallback protection if any are compromised.

What has shifted is the pace of classical cryptographic attacks. A related proposal, BIP-361, warns that the migration window is narrowing, citing improvements in attack efficiency of up to 20 times. Anthropic’s findings reinforce this trend, showing how AI can accelerate such advances.

For HAWK’s smallest parameter set, the estimated cost of recovering a key dropped from about 2^64 operations to 2^38. While larger keys remain difficult to break, increasing key sizes would remove much of HAWK’s performance advantage.

Anthropic disclosed the vulnerability to HAWK’s developers in June and coordinated its release with NIST.

In a separate result, researchers improved attacks on a weakened version of AES—the standard encryption used to secure wallet files—by factors of 200 to 800.

More reassuringly, the model achieved only modest gains—under tenfold—against Poseidon, a hash function widely used in zero-knowledge proof systems for rollups and privacy protocols.

Initially, Claude declined to tackle the AES challenge, describing it as genuinely difficult. After a few prompts, however, it generated billions of tokens over three days and identified the improvement. Verifying the result took nearly a month of work by two researchers.

The timing is striking. This week, the privacy-focused network Zcash rolled out an upgrade introducing a new shielded pool designed to remain secure in a quantum future.

The cryptographic systems meant to carry the industry through that transition are now being tested by AI tools that move faster than the humans who created them.