Bitcoin’s Quantum Strategy Faces Test as AI Cracks Key Algorithm in 60 Hours

The HAWK system has not been deployed anywhere, and Bitcoin does not rely on it. However, it is part of the same race toward quantum-resistant technology that Bitcoin developers are urgently working to address.

Earlier this week, AI company Anthropic revealed that its Claude Mythos Preview model identified a vulnerability that effectively reduces the key strength of HAWK—a proposed alternative to current digital signature systems used in online banking and web payments—by half.

This discovery required around 60 hours of work and approximately $100,000 in computational resources, targeting an algorithm that had already undergone two years of development and multiple rounds of expert human review.

Digital signature schemes like HAWK are designed to verify that a message or dataset originates from the holder of a specific digital key. Bitcoin uses such signatures for every transaction, and they are also fundamental to secure web browsing, as indicated by the padlock symbol in browsers.

Existing signature systems are expected to become vulnerable once quantum computers mature, and HAWK is one of several candidates being evaluated as a replacement. However, it has not yet been deployed in any real-world setting.

Importantly, this research does not impact Bitcoin or Ethereum at present, as both rely on elliptic curve cryptography, which was not targeted in these attacks. Moreover, HAWK is not among the schemes Bitcoin plans to adopt.

The proposal known as BIP-360 outlines a transition to quantum-resistant Bitcoin addresses using three algorithms already standardized by NIST. It includes multiple options to ensure fallback solutions if any single method is compromised by future advances.

What has changed is the pace of progress in classical cryptographic attacks. A related proposal, BIP-361, suggests that the window for transitioning is shrinking, citing improvements in attack efficiency of up to 20 times. Anthropic’s findings support this trend, demonstrating similar acceleration driven by AI models.

For HAWK’s smallest configuration, Anthropic reported that the computational effort required to recover a key dropped from approximately 2^64 operations to 2^38. While larger key sizes remain resistant, increasing key lengths to compensate would undermine many of HAWK’s original advantages.

Anthropic shared its findings with HAWK’s developers in June and coordinated the disclosure with NIST’s public mailing list.

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

Encouragingly for crypto developers, the same model achieved only modest gains (less than tenfold) against Poseidon, a hash function widely used in zero-knowledge proof systems that support rollups and privacy-focused protocols.

Initially, Anthropic’s Claude AI declined to tackle the AES problem, describing it as genuinely difficult with no obvious shortcuts. However, after a few prompts, it generated billions of tokens over three days and ultimately identified the improvement. Verifying the result took two researchers nearly a month.

The timing is notable. On Tuesday, the privacy-focused network Zcash implemented an upgrade introducing a new shielded pool designed to remain recoverable in a future with quantum computers.

The cryptographic systems intended to support this transition are now being rigorously tested by AI tools that can operate faster than the human experts who designed them.

About The Author

ctleditor

See author's posts