Alice & Bob: RSA-2048 could fall to a 22,593-qubit quantum computer

The drop comes mostly from so-called cat qubits, which use superpositions of the electromagnetic field in superconducting resonators. They suppress one type of error, so far fewer extra qubits are needed for error correction. Add more efficient LDPC-cat codes and improvements to Shor's algorithm, and the hardware bill shrinks. A faster variant uses 610,530 qubits spread over ten machines and would finish in about 75 minutes.

Is the 22,593-qubit figure realistic?

Treat it with caution. You can't compare it directly with earlier estimates, because the models use different assumptions and architectures. The authors assume an extremely low error rate of one particular kind, roughly one error per qubit every hundred years. If it happened once an hour, the estimate rises to about 45,000 qubits. Cooling, control and speed requirements come on top. The study is a theoretical analysis that has not been peer reviewed, and Alice & Bob has an obvious interest in showing that its approach has a future.

Should banks be worried?

RSA-2048 is not the same thing as all bank encryption. Modern connections also use symmetric encryption such as AES, plus elliptic-curve key exchange. Breaking RSA does not automatically open someone's account. Shor's algorithm does threaten elliptic-curve cryptography too, though.

Defences exist. In 2024 the US NIST approved post-quantum standards, including ML-KEM for key exchange and ML-DSA for digital signatures. They need no quantum computer. They do need rollout, system updates and retiring old solutions, which in large financial infrastructure can take years.