Research
What we ran, and what did not hold.
We run quantum circuits on IBM hardware to understand the threat we help clients prepare for. Some results held. Several did not, and we corrected them in public.
Shor-ECDLP runs that support a quantum key break
Shor-ECDLP, on toy curves
Ten runs. No quantum break.
We ran Shor’s algorithm for the elliptic-curve discrete logarithm against small curves, 4 to 21 bits, far below real key sizes. 2 runs were full Shor circuits (4 and 6 bits). The other 8 were hybrids, where a classical step did most of the work. The 6 and 8-bit results were computed on a simulator after a code bug. In every run, the key was recovered classically, not by the quantum computer.
Bell test, 2026-07-30
A textbook result, reproduced on hardware.
A Bell test checks whether two measured qubits are correlated more strongly than any classical system allows. The CHSH score S can reach at most 2 for a classical system and about 2.828 for a quantum one. On ibm_marrakesh we measured S = 2.733 ± 0.023, about 32 standard deviations above the classical limit, from 4096 shots at each of four settings. A three-qubit version (GHZ-Mermin) gave M = -3.859 ± 0.008, against a classical limit of 2.
These circuits ran on IBM hardware and the device behaved quantum mechanically. We re-read the counts IBM stores for the first two CHSH jobs and they match the published correlations.
Anything new. This is a known result reproduced on a single chip, and it is not loophole-free: the locality and detection loopholes remain open. It says nothing about our other software.
Corrections
What we claimed, and what we say now.
Each row comes from our public claims registry. The biggest single correction: results we published from 18 and 19 March 2026 were computed on a simulator, not the hardware they were labeled with. Read that correction
Shor-ECDLP circuits executed on real IBM quantum hardware (11 of 12 jobs on ibm_fez) across ten prime-order curves, 4-bit to 21-bit.
Real-hardware execution, every job ID listed on our ledger. Two of the ten runs were full Shor circuits; eight were hybrids where a classical step did most of the work. None yielded a cryptographic break; key recovery was classical. 0 of 10 support a quantum break. Correction, 18 September 2026: the published 6-bit and 8-bit results were computed on a simulator after a code bug; the other eight came from the hardware. The 0 of 10 result is unchanged.
E8 quantum phase estimation (15 qubits, 128 bins), run on an older 133-qubit IBM machine.
The "fine-structure constant detected" reading was retracted: the histogram was a DC noise pile-up measured against the wrong null. The circuit ran; no eigenphase was resolved. Correction, 18 September 2026: the job ran on IBM hardware, but the published value was computed on a simulator after a code bug; see /research/quantum.
E8 Cartan-evolution quantum walk (8 qubits) on real hardware.
The "alpha detected" flag was retracted: the artifact recorded a detection while its hit list was empty. The walk executed; no eigenphase was resolved. Correction, 18 September 2026: the job ran on IBM hardware, but the published value was computed on a simulator after a code bug; see /research/quantum.
A 16-qubit entanglement (GHZ) circuit on real IBM hardware.
Withdrawn. The earlier fidelity 1.0000 claim was retracted in May 2026, and on 18 September 2026 we found no record of this run in our run log and no IBM job we can attribute to it. We no longer claim it ran on hardware.
Keyed HMAC seeded by entropy from a real-hardware circuit run.
The "Bell-certified gate" framing was retracted. The defensible claim is an HMAC seeded by hardware-run entropy; no nonlocality certification. Correction, 18 September 2026: the job ran on IBM hardware, but the published value was computed on a simulator after a code bug; see /research/quantum.
E8 gauge decomposition: three circuits (E8, A8-chain control, random-graph control) submitted to IBM, 8192 shots each.
Entropy and KL divergence are statistically indistinguishable across all three; a calibrated null shows the hit counts are consistent with hardware noise. Not confirmed, not fully retracted (raw per-bin data was not archived, so one alternative could not be tested). Correction, 18 September 2026: the job ran on IBM hardware, but the published value was computed on a simulator after a code bug; see /research/quantum.
Using this work
The same discipline, applied to your cryptography.
IBM job pages are visible only inside our account, so the job IDs we list are a record we cannot edit rather than pages you can open. For researchers, we welcome questions, replication and co-authorship.