Bitcoin's quantum migration plan has always rested on a simple assumption: some of today's cryptographic algorithms will inevitably break. The only question was how fast. A new result from Anthropic's Claude Mythos Preview model suggests the answer may be sooner than expected. In about 60 hours, the AI system found a flaw in HAWK, a digital-signature scheme that had been proposed as a successor to the algorithms protecting modern web and banking systems in a post-quantum era. The discovery reduced the work needed to crack HAWK's smallest key setting from roughly 2^64 operations to 2^38 operations — a thunderous collapse in security margin.
HAWK is one of several candidates that cryptographers have put forward to replace today's signature schemes, such as RSA and elliptic curve cryptography, which would be vulnerable once large-scale quantum computers mature. The scheme was designed to offer robust security with compact keys and signatures, making it attractive for constrained environments. It has never been deployed in production, and Bitcoin does not use it. Nevertheless, the result is being read as a warning: AI-assisted cryptanalysis is advancing quickly, and even carefully studied proposals can harbor hidden weaknesses.
What is HAWK and why does it matter?
HAWK belongs to a new generation of digital signatures built to survive the arrival of quantum computers. Classical digital signatures rely on mathematical problems that are hard for conventional computers but potentially trivial for quantum machines. Shor's algorithm, for example, can efficiently solve the integer factorization and discrete logarithm problems underpinning RSA and elliptic curve cryptography. Once a sufficiently powerful quantum computer exists, it could derive a private key from a public key, breaking encryption and signatures at scale.
To prepare for that day, the cryptographic community has been running standardization processes to select post-quantum algorithms. HAWK was pitched as a candidate in that effort, aimed at delivering strong security while keeping key sizes and signature lengths manageable. Its designers claimed it could provide a practical drop-in replacement for the signatures used in web browsing, banking, and potentially blockchain networks. But the new attack calls those expectations into question, at least for the smallest parameter set.
The attack did not require a quantum computer. It used conventional computing resources powered by AI. Anthropic's Claude Mythos Preview model identified a structural weakness that classical cryptanalysis could exploit. The result lowered the security level from 2^64 operations, which is generally considered borderline but sometimes acceptable for low-risk, short-term use, down to 2^38 operations. That lower number is dangerously small; a determined attacker with enough hardware could potentially mount such an effort in a matter of hours or days, rather than centuries.
An expensive but remarkably fast break
The exact details of the attack are still emerging, but the reported cost paints a clear picture. The team behind the effort spent about 60 hours and $100,000 in computing resources to find the vulnerability. In the context of cryptography, that is both fast and cheap. Traditional cryptanalytic breakthroughs often require years of manual effort by specialized researchers, or massive academic collaborations. An AI system being able to do this in less than three days signals a shift in how cryptographic flaws may be discovered in the future.
The original report also noted that the attack made larger, compensating key sizes far less attractive. In many post-quantum schemes, one way to blunt a discovered weakness is to increase key sizes or signature sizes. But if the flaw scales poorly, or if the compensating sizes undermine the scheme's practical advantages, the candidate loses its appeal. HAWK's selling point was its efficiency. If that efficiency vanishes once the parameters are inflated, then its status as a viable post-quantum replacement is in doubt.
It is important to repeat that HAWK is not deployed anywhere. Bitcoin does not use HAWK, nor do major banks or web services. The attack is not a live exploit. There is no immediate risk to funds in existing Bitcoin or ether addresses. But the episode is a reminder that standardization is not a one-time event. Algorithms must be scrutinized continuously, and AI is becoming a powerful new tool for finding flaws.
Bitcoin's quantum problem is not new
Bitcoin has been aware of the quantum threat for years. The protocol currently uses digital signatures based on elliptic curve cryptography, specifically ECDSA and Schnorr signatures. Under ordinary circumstances, these algorithms are considered secure against classical computers breaking them in any reasonable timeframe. But a fully functional quantum computer running Shor's algorithm could theoretically derive a private key from a public key. That would allow an attacker to spend bitcoin from any address where the public key has been exposed.
Bitcoin addresses in common use are derived from a public key, but many addresses are presented to the network only as hashes. This gives an additional layer of protection: if a public key has never been revealed, a quantum attacker would first need to invert the hash function, which is a harder problem. However, whenever a user sends a transaction, the public key is revealed in the signature. That means the window of vulnerability exists from the moment a transaction is broadcast until it is confirmed and the coins move. Some proposals have discussed moving bitcoin to a quantum-resistant signature scheme before large-scale quantum computers become operational.
Bitcoin's roadmap for quantum resistance is not a single dramatic upgrade. It is more of a staged migration. Developers have explored ideas such as new address formats, new signature types, and careful consensus changes. The underlying assumption has always been that some algorithms might break over time, and the network needs to be flexible enough to adapt. The HAWK episode illustrates exactly the kind of scenario that worries cryptographers: an algorithm that looks good on paper may contain hidden weaknesses that only become apparent after concentrated analysis.
AI is accelerating cryptanalysis
The use of AI in cryptography is not entirely new. Researchers have previously used machine learning to find weaknesses in ciphers, to attack public-key encryption, and even to improve side-channel attacks. What is notable about the HAWK result is the speed and cost. A 60-hour effort with a $100,000 budget is accessible to well-funded companies, nation-states, and even sophisticated criminal groups. That changes the threat model.
In the past, breaking a cryptographic scheme required deep mathematical insight and often a stroke of genius. Now, an AI model can help scan the structure of a scheme and identify patterns that humans might miss. This does not mean every algorithm will fall overnight. But it does mean that the margin for error in cryptographic design is shrinking. Standards bodies, blockchain developers, and security companies will need to adopt a more agile mindset, continuously reviewing and replacing algorithms before they fail.
The HAWK attack also raises questions about the broader ecosystem of post-quantum cryptography. Several candidate schemes have already been submitted to ongoing standardization efforts. If one can be weakened in 60 hours, others may be vulnerable as well. Cryptographers have long warned that these early candidates need multiple independent analyses. AI, despite its power, is simply one more tool in the cryptanalyst's toolkit. Still, the ability to automate vulnerability discovery represents a genuine leap in what is practically feasible.
What this means for blockchain networks
Bitcoin and other blockchain networks face a unique challenge. Unlike a web server that can update a library or a certificate quickly, a decentralized network requires consensus. Changing the underlying signature scheme requires every node, wallet, and exchange to coordinate. That process can take years, and it must be planned long before a real quantum emergency. The HAWK result is a reminder that cryptographers cannot afford to be complacent.
Several projects have already begun experimenting with quantum-resistant signatures. Some have proposed using hash-based signatures such as Lamport signatures or stateful schemes like XMSS, though those come with their own trade-offs. Others are looking at lattice-based schemes, multivariate cryptography, or other families. Bitcoin, with its conservative culture, will likely take a careful path. The community has historically prioritized stability and security over speed. But the HAWK attack suggests that the timeline for quantum resistance may be shorter than once thought, at least in terms of cryptanalytic progress.
The $100,000 cost of the attack is also significant. It is low enough that it could be repeated by many actors. If AI systems continue to improve, the cost of finding new weaknesses may drop even further. This creates a paradoxical situation: quantum computers may still be years away, but the classical tools used to attack post-quantum candidates are improving rapidly. The community must prepare for a world where algorithms can be broken not just by a future quantum machine, but by a sophisticated AI using ordinary silicon.
For now, bitcoin and ether remain safe. ECDSA and Schnorr signatures are still believed to be secure against all known classical attacks. The HAWK scheme is not part of Bitcoin's codebase, and no funds are exposed. But the episode is a valuable stress test for the broader ecosystem. It shows that AI can act as an accelerant in the cryptanalysis world, and that even the most promising algorithms need constant scrutiny.
The road to quantum-resistant blockchains will be long and arduous. It will require not only new algorithms, but also new processes for evaluating them. The HAWK attack is a reminder that the threat is not only quantum. It is also the rapidly improving ability of AI to find flaws in today's best cryptographic designs. Bitcoin's quantum plan may assume that some algorithms break; the only real question is how soon and how badly. A 60-hour break of a candidate signature scheme is a sobering data point in that ongoing race.
Source: Coindesk News