I think the FCA’s Quantum Experimentation Group will be judged on its dataset long before anyone judges its qubits. The regulator has given the group a synthetic financial crime dataset and a comparison against classical methods, and those two choices decide whether the results will mean anything to a bank’s fraud team.
The hardware gets the headlines
The FCA announced the group on 5 November 2026. It will bring together financial services firms, quantum technology providers and national research laboratories, launch in the first quarter of 2027, and start with quantum machine learning for detecting money laundering and fraud. Most coverage will focus on the word quantum.
The detail that matters more sits in the second paragraph of the release. Participants will use an FCA synthetic dataset, which the release describes as artificially generated data, to overcome what it calls a key barrier to testing: access to suitable data. The FCA’s page for the group repeats the point: “Suitable data also remains a practical barrier to experimentation.”
Advertisement
Why a common dataset is the right design
Quantum proofs of concept in finance have mostly run on whatever data a research team could get. The FCA’s October 2025 research note says many quantum teams sit within innovation units with limited access to sensitive financial data, which forces reliance on open or synthetic datasets. Each team then reports its own accuracy figure on its own data, and no two figures can be set side by side.
One dataset and one defined problem change that. The group page says participants will use them to develop, test and evaluate quantum approaches, including comparisons against classical techniques. A result on a shared benchmark can be checked by the next participant. A result on private data can only be believed or doubted.
Jessica Rusu, chief data, information and intelligence officer at the FCA, put the aim in terms of common problems: “By bringing together firms, technology providers and researchers we can build readiness and identify practical applications for common industry challenges, starting with financial crime.” A common challenge needs a common yardstick, and the dataset is that yardstick.
The strongest objection
The obvious objection is that synthetic data is generated, so it may leave out the messy patterns that real mule networks and real fraud rings produce. A model that does well on a tidy simulation could fall apart on live transactions. That is a fair worry, and the FCA’s materials do not say how closely the dataset will track real behaviour or how it was built.
I still think the shared set is the better choice, for two reasons. First, the alternative on offer to most quantum teams is no realistic data at all, or data too sensitive to share across firms, providers and labs in one group. Second, the first question the group can answer is a narrow one: does a quantum approach match or beat a classical baseline on the same problem. That question can be answered on a synthetic set even if the answer will not transfer to production unchanged. A negative result on a clean benchmark would be informative, because a method that cannot beat classical models on simplified data has little prospect on harder data.
What firms should ask for
If I were a head of financial crime deciding whether to register by 31 December 2026, I would want three things settled before the first cohort session. I would want the classical baseline defined up front, with its detection and false-positive rates published to participants. I would want the FCA to say what evidence it would accept that a quantum model’s output holds up and stays stable, since its own research note says supervisors require clear benchmarks and evidence of that kind, and that quantum reproducibility will be complicated by probabilistic outputs and hardware variability. And I would want to know how much of each participant’s learning the FCA will share, because the group page says only high-level learning will be shared.
None of that is in the announcement yet, and none of it needs quantum expertise to ask for. The group launches in about five months, and the firms that raise these points during registration will shape the first problem statement that the FCA writes.
Source: FCA press release, “FCA to explore quantum computing to tackle financial crime”
