Artificial intelligence could one day threaten the cryptographic foundations of Bitcoin before quantum computers ever become powerful enough to break today's signatures. That is the provocative possibility raised by a new development from AI research laboratory Anthropic, which says one of its frontier models has discovered novel attacks against weakened versions of established encryption algorithms. The finding has reopened a discussion about whether the blockchain industry is preparing for the right existential threat.
Bitcoin currently depends on ECDSA, a class of elliptic curve cryptography that has served as the backbone of transaction validation for over a decade. Cryptographers have long warned that a sufficiently capable quantum computer could eventually solve the mathematical problems underlying ECDSA, allowing someone who obtains a public key to forge a signature and spend funds. That risk has motivated a gradual movement toward post-quantum signature schemes, with lattice-based cryptography emerging as one of the leading candidates for a future upgrade.
According to Anthropic, its Claude Mythos Preview model has discovered previously unknown attacks against weakened versions of the Advanced Encryption Standard, better known as AES. AES is one of the most widely used symmetric encryption systems in the world, protecting everything from internet traffic to stored data. The AI-generated attack, the company claimed, made breaking the weakened algorithm between 200 and 1,000 times faster than prior approaches. Anthropic also reported that the model devised an improved attack against HAWK, a post-quantum cryptographic signature scheme that has been considered as a possible component of future quantum-safe security infrastructure.
The announcement was not framed as an immediate danger to deployed cryptography. The targets were deliberately weakened algorithms used for research, not full-strength production implementations. Anthropic stated that modern AES remains secure. Yet the fact that a language model produced original cryptanalytic results has generated broader discussions about how quickly AI systems are improving at mathematical discovery and what that means for the transition to quantum-resistant systems.
Advancing cryptographic research
The discovery is being described as a demonstration that frontier AI models are becoming capable of producing original cryptographic research. This is a significant milestone in a field where most breakthroughs still require years of human expertise and intuition. In the past, machine learning models were mostly used to assist cryptanalysts by classifying or optimizing certain parts of an attack, but the actual creative insight remained human. The Anthropic model appears to have crossed a threshold by generating new attack paths on its own.
Cryptographic research is a special domain because it involves both constructing codes and breaking them. Historically, progress has come from a close study of algebraic structures, combinatorial properties and computational complexity. AI models trained across massive datasets can discover patterns and correlations that might not be obvious to human researchers. That strength is especially relevant when searching through enormous spaces of possible attack configurations.
The results reported by Anthropic are still preliminary and have yet to be independently verified or published in a peer-reviewed venue. However, the company says the attacks have been validated by its internal researchers. If confirmed, they would mark a notable step toward using AI not only as a tool for writing code or assisting with mundane tasks, but as an autonomous contributor to security research.
Post-quantum cryptography has become a priority because quantum computers pose a specific threat to public-key algorithms. Shor's algorithm, first proposed in 1994, describes how a sufficiently large quantum computer could efficiently factor large integers and compute discrete logarithms. Many of the cryptosystems used today, including ECDSA, rely on problems that Shor's algorithm could solve. A quantum computer with enough stable qubits and error correction could theoretically recover the private key from a Bitcoin public key, which is exposed whenever a transaction is made.
There is also Grover's algorithm, which gives a quadratic speedup for brute-force searches. For AES, this means that a quantum computer could reduce the effective key strength by half, making a 128-bit key roughly as difficult to crack as a 64-bit key. This is a reason that post-quantum recommendations often include larger symmetric key sizes or a move to 256-bit AES. Yet AES is generally considered less vulnerable than public-key systems because Grover's algorithm does not completely break it; it merely reduces the security margin.
Why it matters for Bitcoin
Bitcoin's vulnerability to quantum computing lies primarily in its use of ECDSA. When a user creates a transaction, they reveal their public key. An attacker with a powerful quantum computer could, in principle, use that public key to compute the private key and then construct a spending transaction from the victim's address. This is often described as a 'harvest now, decrypt later' problem, although in the case of public-key cryptography the decryption step can be applied retroactively to exposed keys once a sufficiently large quantum computer is available.
Bitcoin developers have been discussing potential upgrades for years. One widely discussed idea is to implement a post-quantum signature scheme, with lattice-based cryptography as a leading option. Lattice problems are believed to be hard for both classical and quantum computers, and many cryptographic systems have been built around them. The NIST competition that selected several post-quantum algorithms included lattice-based schemes for digital signatures and key encapsulation. Some of these have already been incorporated into protocols such as Signal's quantum-resistant extension, but deploying them in Bitcoin would be a major undertaking.
The challenge is not simply choosing a new signature algorithm. Bitcoin has an enormous amount of value locked in legacy addresses, and any migration must be backward compatible. A fork that changes the signature scheme would need broad consensus among users, miners and developers. Even if a post-quantum algorithm is selected today, the process of writing a Bitcoin Improvement Proposal, testing, deploying and activating it could take years. That timeline is one reason the cryptocurrency community has watched AI research with growing concern.
The new Anthropic findings complicate the picture. Lattice-based schemes like HAWK are not yet widely deployed, but they are part of the post-quantum toolbox. If an AI system can improve attacks against such schemes, even against weakened versions, it raises the possibility that the future cryptographic environment will be more complex than a simple race between classical computers and quantum machines. An advanced AI model, running on classical hardware, could potentially find structural weaknesses in algorithms that were designed to resist quantum attacks.
No immediate threat
Researchers were quick to stress that neither Bitcoin nor today's internet encryption is in immediate danger. The attacks reported by Anthropic targeted deliberately weakened versions of AES and HAWK, not the full-strength algorithms used in production systems. Breaking modern AES with a 128-bit key through brute force would still require an astronomically large number of attempts. The same is true for the standardized versions of post-quantum signatures currently under consideration.
The speedup of 200 to 1,000 times is significant in a laboratory setting, but it does not translate to a practical break. Cryptographic security margins are designed with a huge factor of safety, and a thousandfold improvement is still far from enough to threaten a system with 128-bit security. Even a millionfold improvement might not suffice, because the search space is so vast. The discovery is thus more valuable as a sign of AI's potential than as an immediate weapon against real-world systems.
There is also a difference between discovering an attack on a weakened variant and breaking a full-strength cipher. Researchers often begin by analyzing simplified versions to understand structural properties. Those insights can sometimes be extended to the full version, but that is by no means guaranteed. The history of cryptography contains many examples of attacks on reduced-round versions of ciphers that never scaled to the complete algorithm.
AI and quantum: a dual challenge
If AI systems continue to accelerate mathematical discovery, developers of Bitcoin and other blockchain networks may eventually have to evaluate post-quantum security against both quantum and AI cryptanalysis. This is not a purely hypothetical scenario. The rate of progress in language models has been rapid, and there is no reason to assume that cryptanalytic capability will not follow.
The idea of an AI model breaking a weakened version of AES has a certain irony for the cryptocurrency community. Bitcoin is often described as a technology that combines cryptography with economic incentives. Its security model is based on the assumption that certain mathematical problems are hard. That assumption may one day be challenged by a machine that can explore mathematical structures in ways no human has imagined. The 'funniest timeline,' as one industry observer put it, would be one in which Bitcoin upgrades to a lattice-based post-quantum signature scheme only for advanced large language models to break the underlying mathematics before quantum computers ever manage to pose a threat.
For now, the recommended course for Bitcoin stakeholders is not panic, but attention. Cryptographic research powered by AI is likely to become more common, and the same tools that discover attacks can also be used to design and verify new defenses. The development of quantum-resistant cryptography has always been a moving target; the addition of AI as an active participant in cryptanalysis makes it even more important to build robust, well-tested standards and to avoid complacency.
The question of whether AI can beat quantum computers to the punch is not answerable with certainty. What is clear is that Anthropic's discovery has added a new variable to Bitcoin's long-term security equation. The race to prepare for a future with powerful computing systems, whether quantum or AI-driven, is no longer purely theoretical.
Source:U.Today News