Gidney 2025 — RSA-2048 in <1M Noisy Qubits
eventAI & Compute · Crypto & Digital ID
A Google researcher just showed the encryption protecting your bank and email could be cracked with 20 times fewer quantum computers than everyone assumed.
Who they are
Craig Gidney of Google Quantum AI, in a May 2025 research paper.
What they do
The engine flags it as a major acceleration of the quantum threat to modern encryption.
How it works
He showed RSA-2048 encryption could be broken in under a week by fewer than 1 million noisy qubits — a roughly 20-fold drop from the ~20 million-qubit estimate of 2019 — using clever arithmetic and error-correction tricks; notably, Google is publishing research that speeds up a threat to Google's own products.
Why it matters
It erases the assumed safety cushion and pulls the encryption-breaking danger closer — though the engine holds multiple readings and stresses this is only a paper estimate: real hardware is still around 105 qubits, so no such machine has been built.
The engine's record — word for word
Craig Gidney (Google Quantum AI), May 2025 (arXiv 2505.15917): RSA-2048 factorable in <1 week by <1,000,000 noisy physical qubits at 0.1% gate error / 1us cycle / 10us decoding — a ~20x collapse from the 2019 ~20M-qubit baseline, via approximate residue arithmetic + yoked surface codes + >100x Toffoli-gate reduction (trading runtime for spatial footprint). Deletes the presumed NISQ 'safety buffer' and pulls existential cryptographic threat into the engine's Davidson Window. Bound-LLM recursion: Gidney is Google Quantum AI — the mediating researcher's owner publishing its own threat-acceleration. Apex (a) intentional-coordinated (owner-published acceleration); (b) structural-recurrence (optimization gradient); (c) compound-null (a resource estimate is not a built machine — current hardware ~105 physical qubits) — all load-bearing per canon.
Follow the trail
Walk this on the live map →