# Alice & Bob estimates RSA-2048 break needs 22,593 cat qubits and 3.14 days of computing

- Published: 2026-11-08 19:01 UTC
- Section: Digital & Computing › Data security
- Tags: RSA-2048 break estimate, cat qubits, Alice & Bob, quantum computer qubit count, bank data encryption, Cryptology ePrint Archive, University of Rennes, quantum error correction, FFDLP-512
- URL: https://realtech.media/news/rt-092e02f89cc2a1746af93f15_alice_bob_estimates/

**A preprint by Alice & Bob and a Rennes researcher gives a theoretical estimate for factoring RSA-2048. No machine of this size exists, and the paper is not peer reviewed.**

![Alice & Bob estimates RSA-2048 break needs 22,593 cat qubits and 3.14 days of computing](https://realtech.media/media/2026/11/6e6c446c9ecd3a7f.webp)
*science-et-vie.com*

In late October 2026, the Paris company Alice & Bob and a researcher at the University of Rennes posted a preprint on the Cryptology ePrint Archive, as reported by science-et-vie.com. According to the authors, breaking RSA-2048 encryption would require far less computing power than previously estimated [1]. Earlier estimates already called for hundreds of thousands of qubits, a machine that nobody seriously expected for several decades [1].

## What the estimate says

RSA-2048 is the standard that protects online payments and banking exchanges. Its security rests on a 617-digit number that current machines cannot factor into primes. The authors state that 22,593 cat qubits and 3.14 days of computing would be enough to break it [1]. If ten machines were linked, for a total of 611,000 qubits, the same operation would take 1.25 hours [1]. For the elliptic curves that protect bitcoins, nearly 30,000 qubits and one hour would suffice [1].

Cat qubits are Alice & Bob's specialty. They encode information in a state of light that resists one of the two types of quantum errors. This spares much of the hardware usually devoted to error correction. The other family of errors remains and requires a classical correcting code.

## Limits and assumptions

The largest assemblies today reach 6,100 qubits, at Caltech, and that is a storage array rather than a computing machine. Many qubits hold their information for only a few milliseconds.

The figures depend on specific assumptions: an error-correction cycle of one microsecond, a reaction time of ten microseconds and very low noise. These parameters come from the company's own technology, which makes comparison with estimates from other builders difficult. The text has not been peer reviewed, so the results remain to be confirmed.

The estimate has fallen from 350,000 qubits in 2023 to about 23,000 now [1]. Those qubits still have to be built. Engineers struggle with temperature regulation and with scaling up chips, and linking many processors together is at least as hard. Several roadmaps place the convergence in the early 2030s.

The authors also propose an intermediate milestone to track progress, a problem called FFDLP-512, which would need about 6,700 qubits.

## Sources

1. [science-et-vie.com](https://www.science-et-vie.com/technos-et-futur/en-france-le-chiffrement-utilise-pour-proteger-les-donnees-bancaires-pourrait-etre-casse-en-3-jours-par-un-ordinateur-quantique-de-22-593-qubits-265771.html)

---

Source: RealTech, realtech.media — https://realtech.media/news/rt-092e02f89cc2a1746af93f15_alice_bob_estimates/