Anthropic's artificial intelligence (AI) model Claude autonomously designed eight types of computational procedures for quantum computers. In seven of them, it reduced the number of qubits and operations needed compared with existing optimal designs. Some designs improved resource efficiency by more than 10 times. Interest is also focusing on how higher computational efficiency in quantum computing could affect the security systems of cryptocurrencies such as Bitcoin.

CoinPost, a blockchain media outlet, reported on Oct. 17 local time that Anthropic disclosed the research results the previous day. The company said Claude designed eight quantum computation procedures and reduced resource consumption in seven of them compared with existing methods. Reducing both the number of qubits and the number of operations needed is important to make quantum computers more practical.

The research is drawing attention because making quantum computation more efficient could also affect the amount of resources needed for codebreaking. Bitcoin and Ethereum use elliptic-curve cryptography for transaction signatures. With conventional computers, it is difficult to derive a private key from a public key within a realistic timeframe, but that security assumption could be shaken if sufficiently large quantum computers emerge. If a private key is exposed, attackers could gain the possibility of moving assets protected by that key.

Anthropic focused in particular on the 'point addition' operation, which is repeatedly used in quantum computations related to elliptic-curve cryptography. It is the process of calculating a new point using two points on an elliptic curve. Because it is an important operation in quantum algorithms needed for codebreaking, reducing the resources required to perform it could also affect overall computational efficiency.

The researchers presented Claude with specifications of computational procedures in natural language and built a program that verified the accuracy of designs and evaluated them based on the product of the number of qubits and the number of operations. Claude repeatedly submitted designs and improved them based on the evaluation results. Instead of humans directly designing detailed computational procedures, the approach had AI use verification and scoring results to search for more efficient methods.

For tasks related to codebreaking computations, it explored new calculation methods for the costly point addition operation for about 53 hours with almost no human intervention. One of the resulting designs required 12 percent more qubits than the existing minimum design, but reduced the number of operations to about one-seventeenth. It found an optimisation path that slightly increased qubit use while sharply lowering the overall operational burden.

The research does not mean it actually broke the cryptographic systems of Bitcoin or Ethereum. Anthropic said in the announcement it did not carry out attacks directly targeting cryptocurrencies and did not present the total size of the quantum computer needed to break encryption. Improving the efficiency of computational procedures and implementing a quantum computer capable of actually breaking encryption are separate issues.

The current scale of quantum computers is also still cited as a major constraint. Explanations presented in the material say today's quantum computers remain at the level of hundreds to thousands of qubits. Google previously disclosed an estimate in March that the qubits needed to break Bitcoin's signature cryptography could be fewer than 500,000. The actual requirement can vary depending on factors including the calculation method and error correction, limiting simple comparisons.

Anthropic also suggested the possibility that AI could move beyond optimising individual computational procedures for quantum computers to jointly optimising components and even taking charge of designing entire systems tailored to specific problems. It said AI could expand its role from improving software-level computational algorithms to broader quantum computing hardware design.

The result is closer to an example showing that AI can improve the efficiency of quantum computation than to an immediate threat to cryptocurrency security. Still, if AI-based optimisation continues, it is expected to influence discussions about the pace of quantum computer practicalisation and long-term security responses for cryptographic systems.