Anthropic Claude AI scientific breakthroughs

Anthropic’s Claude AI models have demonstrated significant advancements in solving complex scientific and engineering problems. In late September 2026, the Claude Science system successfully calculated nine-loop scattering amplitudes in N=4 super Yang-Mills theory, surpassing previous human records in theoretical physics.

By mid-October 2026, the Claude Mythos 5 model achieved a breakthrough in quantum computing through a collaboration with Harvard University and QuEra Computing. In studies covering eight computational tasks—including chemistry, physics simulations, and number theory—the AI developed circuits that outperform existing benchmarks by reducing required qubits and gates.

A notable achievement included a design for elliptic-curve operations that reduced gate requirements by 17 times. Additionally, for the Hermitian-DQI decoder, the model reduced combined resource costs by more than 9,100 times compared to previous baselines. While these improvements were verified through simulations and mathematical proofs, testing on physical quantum hardware remains pending.

On October 18, 2026, further reports detailed how Claude autonomously improved quantum circuit designs, specifically impacting cryptography. In designing circuits for point addition in elliptic curve cryptography—used in digital signatures for cryptocurrencies like Bitcoin and Ethereum—the AI reduced the number of high-cost Toffoli gates to approximately ‘one-seventeenth’ of previous requirements. This was achieved by discovering mathematical rules to reconstruct processing history, allowing for efficient computation without storing extensive intermediate data. Furthermore, the AI improved the Decoded Quantum Interferometry (DQI) algorithm, reducing the Toffoli gate count to about ‘one-twelfth’ of Google’s previous design. While not an immediate threat to current security, these findings suggest the practical application of quantum computing may accelerate the transition to quantum-resistant cryptography.