Anthropic AI improves quantum circuit designs for cryptography

Anthropic has reported that its AI model, Claude, has autonomously improved the design of several quantum circuits, outperforming existing best-known designs in some instances by more than tenfold. The research focused on eight types of fundamental circuits used in fields such as chemistry, optimization, and number theory.

A significant finding involves the design of circuits for point addition in elliptic curve cryptography, a method used in digital signatures for cryptocurrencies like Bitcoin and Ethereum. While the new design uses approximately 12% more qubits than the previous minimal design, it reduces the number of Toffoli gates—a high-cost operation—to about one-seventeenth of the previous requirement. Claude achieved this by discovering mathematical rules to reconstruct processing history, allowing for more efficient computation without the need to store extensive intermediate data.

Additionally, the AI improved the decoding process for the Decoded Quantum Interferometry (DQI) algorithm, reducing the required Toffoli gate count to about one-twelfth of Google’s previous design. While these advancements do not pose an immediate threat to current cryptographic security, they suggest that the practical application of quantum computers could arrive sooner than expected, potentially accelerating the transition to quantum-resistant cryptography.