Anthropic has announced that its Claude Mythos Preview model identified a previously unknown mathematical weakness in HAWK, a lattice-based post-quantum digital signature scheme that had already withstood two years of expert review. The research required roughly 60 hours of computation and approximately $100,000 in computing costs. While the discovery does not affect Bitcoin, Ethereum, or the encryption currently protecting internet infrastructure, it demonstrates how AI is emerging as a significant tool for stress-testing the cryptographic systems that may secure tomorrow's digital economy.

The finding is particularly relevant to the post-quantum review process led by the U.S. National Institute of Standards and Technology (NIST). NIST has already published its first set of post-quantum standards—FIPS 203, 204, and 205—but continues to evaluate additional digital signature candidates through supplementary rounds aimed at broadening the diversity of its cryptographic portfolio. HAWK is part of that ongoing effort, and the goal is to identify and eliminate weaknesses in candidate algorithms before they are deployed at scale.

Targeting Tomorrow's Encryption, Not Today's

HAWK is one of several digital signature schemes currently under evaluation through NIST's post-quantum cryptography standardization process. It has never been deployed in real-world financial systems or blockchain networks.

Digital signatures are the mathematical mechanisms that verify authenticity whenever someone sends cryptocurrency, logs into online banking, or accesses a secure website. Today's signature algorithms remain effective against conventional computers, but sufficiently powerful quantum computers could eventually solve the mathematical problems underpinning many current encryption methods—a threat posed specifically by Shor's algorithm, which targets the factoring and discrete logarithm problems behind widely used schemes like RSA and ECDSA. This is why researchers have spent years developing post-quantum algorithms designed to withstand such attacks.

According to Anthropic's research findings, Claude identified a faster method to attack HAWK's smallest challenge parameter, reducing the estimated computational work from roughly 2⁶⁴ operations to 2³⁸. For context, 2⁶⁴ operations is near the boundary of what a well-resourced organization could plausibly achieve with conventional hardware over time, while 2³⁸ falls within reach of modest computational resources. This represents a significant advance in cryptanalysis. However, the algorithm was not broken in any production setting, as it was never protecting real users. Larger security parameters also remain computationally impractical to attack.

No Threat to Bitcoin or Ethereum

Bitcoin secures its transactions using the Elliptic Curve Digital Signature Algorithm (ECDSA), not HAWK. Ethereum similarly relies on different cryptographic methods. The weakness Anthropic discovered has no bearing on either blockchain's current security model.

The research is nonetheless relevant to Bitcoin because ECDSA itself is among the algorithms that Shor's algorithm could eventually compromise on a sufficiently powerful quantum computer, and developers have been actively discussing how and when the network should migrate to quantum-resistant cryptography. One proposal, BIP-360, recommends supporting multiple standardized post-quantum signature algorithms rather than relying on a single replacement. Another, BIP-361, outlines a specific framework for integrating quantum-resistant transaction types. Anthropic's findings reinforce the logic of that diversified approach: if one proposed algorithm later proves weaker than expected, developers would still have other vetted alternatives available.

As The Street reported, the discovery validates Bitcoin's cautious strategy rather than exposing a vulnerability in the network itself.

AI as a Cryptanalytic Tool

Perhaps the most significant takeaway extends beyond Bitcoin or any single blockchain.

For decades, new encryption systems were tested almost exclusively by teams of mathematicians and security researchers. Reviews frequently took years, as every potential weakness had to be explored manually.

According to Anthropic, Claude Mythos initially refused one of the cryptographic challenges, judging the problem too difficult. After additional prompting and several days of computation, the model generated more than one billion output tokens before identifying a novel mathematical shortcut. Human researchers subsequently spent weeks verifying the result before disclosing it to the HAWK team and NIST through a responsible disclosure process.

This workflow may become increasingly common. Rather than replacing cryptographers, AI can rapidly explore mathematical possibilities while human experts confirm whether the discoveries hold up to scrutiny. For organizations like NIST, this could ultimately strengthen the review process, ensuring that future cryptographic standards must withstand not only years of human analysis but also AI-assisted testing before reaching production.

CoinDesk noted that Bitcoin's quantum migration strategy already assumes some candidate algorithms may break, making AI-assisted cryptanalysis a valuable early warning system.

Industry-Wide Quantum Preparations

Bitcoin is far from alone in preparing for a future where quantum computers could threaten today's cryptography. Developers across the blockchain industry have been researching quantum-resistant upgrades, and some privacy-focused networks have already introduced features designed to facilitate smoother future migrations.

Governments and major technology companies are similarly transitioning toward post-quantum standards. Google has previously announced a roadmap to migrate portions of its infrastructure to quantum-safe cryptography, signaling that the issue has moved beyond academic discussion.

Specialist firms such as PostQuantum have also been analyzing AI's emerging role in cryptographic evaluation.

For Bitcoin, the latest research supports a deliberate approach rather than a rushed one. Developers have consistently favored algorithms that have undergone extensive public scrutiny over those that promise speed or convenience without proven resilience.

What Comes Next

Attention will likely focus on three areas in the coming years.

First is the pace at which AI improves at cryptanalysis. Anthropic itself observed that language models have progressed from struggling with basic cryptographic tasks to uncovering weaknesses that eluded years of expert review.

Second is NIST's ongoing evaluation of post-quantum algorithms. Each newly approved standard will likely face increasingly sophisticated AI-assisted testing before widespread adoption.

Third is Bitcoin's quantum roadmap. Discussions around proposals such as BIP-360 and BIP-361 are expected to continue as developers weigh how and when the network should eventually transition to quantum-resistant signatures.

While practical quantum attacks remain a future concern rather than an immediate one, planning years in advance is widely regarded as the most prudent course of action.