A Head of Innovation at a UK fintech has a model that looks promising on a laptop and nowhere to test it properly. The FCA now offers a route for exactly that moment. Its Supercharged Sandbox gives selected firms a secure cloud environment with GPU-enabled infrastructure, curated and synthetic datasets and expert support, at no cost, according to the FCA's Supercharged Sandbox page. Demand is rising: cohort 2 drew 199 applications, a 51% increase on cohort 1, for 21 places, one fewer than cohort 1.
This guide is for the CTO or Head of Innovation deciding whether to apply if the FCA opens another cohort. It covers what the Sandbox offers, how a cohort runs, how it differs from the FCA's other innovation routes, and what it cannot tell you about your build. It deliberately stops at deployment: live monitoring of AI in production is covered in our guide to FCA AI Live Testing evidence, and senior manager accountability sits in our SM&CR AI accountability guide.
This guide is shipped by the team behind DATS, the five-stage AI consulting system from DILR.AI, run by senior practitioners who ship code, not decks. Or see AI operating model design, which covers governance, RACI and lifecycle, audit-ready by design.
What is the FCA Supercharged Sandbox?
The FCA Supercharged Sandbox is a cohort-based programme in which selected firms experiment with, test and develop AI use cases for financial services inside a secure cloud environment. Participants get GPU-enabled compute, enterprise AI tooling, synthetic and curated datasets, mentorship from a network of experts, and a showcase slot at the end. It is free to join and sits in the FCA's AI Lab.
The FCA announced the Sandbox on 9 June 2025 as a collaboration with NVIDIA, giving firms NVIDIA accelerated computing and NVIDIA AI Enterprise software. It builds on the existing Digital Sandbox infrastructure provided by NayaOne. For cohort 2, the FCA added a collaboration with Anthropic, so participants also have access to Claude, including Claude Code and Claude Cowork, according to the FCA's July 2026 cohort 2 announcement.
The programme sits inside the regulator's AI Lab, alongside AI Live Testing, AI Spotlight and the AI Input Zone. The FCA positions it for firms in the discovery and experiment phase, while AI Live Testing serves firms further along and ready to deploy. That split matters for the decision this guide is about, and it is the same boundary our enterprise AI consulting guide draws between a pilot and a production placement.
Three facts frame everything else. There is no cost to take part. A firm does not need to be FCA regulated, provided its solution is relevant to UK financial services or markets. And participation does not indicate FCA approval, endorsement or authorisation of a product or service, which the FCA states twice on the same page.
How competitive is a Supercharged Sandbox place?
A Supercharged Sandbox place is getting harder to win. According to the FCA, 132 firms applied to cohort 1, which ran from September 2025 to January 2026, and 22 were selected. Cohort 2, which runs from July to December 2026, received 199 applications, a 51% increase, and the FCA selected 21 organisations. Applications rose by about half while the number of places stayed almost flat.
The FCA frames the demand in terms of trust. Announcing cohort 2, Jessica Rusu, the FCA's chief data, intelligence and information officer, said: "The high level of interest in the Supercharged Sandbox demonstrates the demand for trusted environments where firms can experiment safely and responsibly." The named participants range from Scottish Widows to TrueLayer and the Money Advice Trust, so a place is not reserved for start-ups.
Cohort 2's use cases cluster into five themes the FCA lists: safer agent-led payments and commerce, fraud and economic crime detection, AI governance and accountability, access for vulnerable and underserved consumers, and compliance and business automation. Fraud is a familiar fintech pressure point, as our fintech AI hub sets out.
How does the Supercharged Sandbox programme work?
The Supercharged Sandbox runs in five stages that the FCA publishes: apply, assessment, cohort onboarding, build and test, and showcase. Firms submit an online application, the FCA assesses it against five criteria, selected firms are onboarded, participants build and test their use case over the cohort period, and firms can then present their work to industry, regulatory and technology stakeholders.
Cohort 2's calendar shows how tight the window is. Applications opened on 5 May 2026 and closed on 1 June 2026, applicants heard on 26 June, the cohort launched on 13 July, the showcase runs on 25 and 26 November, and sandbox access ends on 31 December 2026. A firm that misses a four-week application window waits for any later cohort, so the preparation has to be done before the window opens, not during it.
The support model is deliberately light. The FCA says participants receive documentation, onboarding support and guidance from a network of experts, and that the programme is designed to enable firms to build independently. Participants should come prepared with the expertise needed to develop their solution. The Sandbox lends you a lab; it does not lend you a team. In a production programme, that is the role an embedded build team such as our AI execution office plays, working on production placements the client owns rather than on experiments.
Firms may bring their own data and models, subject to the Sandbox terms of use and technical constraints. That matters more than it first appears, and we return to it below.
How is it different from the Digital Sandbox, Regulatory Sandbox and AI Live Testing?
The Supercharged Sandbox is one of four FCA routes a firm building AI might consider, and they answer different questions. The Digital Sandbox offers data and APIs for early experimentation. The Supercharged Sandbox adds GPU compute and AI tooling for pre-deployment testing. The Regulatory Sandbox supports live tests with real consumers. AI Live Testing supports AI systems already heading into live markets.
| Route | What it is for | When a firm uses it | How a firm joins |
|---|---|---|---|
| Digital Sandbox | 300+ synthetic, public, anonymised and pseudonymised data sets and over 1000 API end points for proofs of concept | Early experimentation; a typical project runs 3 to 12 months | Applications through the Digital Sandbox platform, with a decision aimed for within 1 month |
| Supercharged Sandbox | GPU compute, enterprise AI tooling, synthetic datasets and expert support | Discovery and experiment phase, up to validating components pre-deployment | Cohort applications in a fixed window |
| Regulatory Sandbox | Testing new products live in the market with real consumers | A live test, typically small scale, limited duration and limited consumers | Applications at any point in the year |
| AI Live Testing | Testing AI systems in real-world conditions with regulatory support | A proof of concept built, with deployment and monitoring plans | Cohort applications; the second ran 19 January to 24 March 2026 |
The Digital Sandbox is the permanent foundation the Supercharged Sandbox was built on. If your problem is access to realistic financial data rather than compute, it may answer it without waiting for a cohort.
The Regulatory Sandbox is a different instrument altogether. It is for testing with real consumers, it is not regulatory exempt, and any authorisation a firm gains through it is restricted to the agreed test. Do not confuse its tools, such as restricted authorisation or individual guidance, with anything the Supercharged Sandbox offers.
AI Live Testing expects a firm to have already developed a proof of concept, considered pre-deployment testing and planned post-deployment monitoring. Our AI Live Testing guide covers that route and the evidence it asks for, and the wider map of which regulator expects what is in our sector regulator framework guide. Other regulators run parallel schemes, such as the MHRA's AI Airlock, which we covered in our MHRA AI Airlock analysis.
What does the Supercharged Sandbox not test?
The Supercharged Sandbox tests whether an AI idea works on curated or synthetic data with good compute behind it. It does not test whether the system fits your own processes, your legacy platforms or your real customer data, unless you bring that data in. Those are the non-regulatory constraints UK financial firms now report most often, so a sandbox result is evidence about the model, not about the deployment.
The 2026 Bank of England and FCA survey locates where the friction has moved. Integration into existing business processes and decisions was the most reported large non-regulatory constraint, cited by 23% of respondents, followed by legacy systems or infrastructure at 22%, according to Artificial intelligence in UK financial services 2026. In the 2024 survey the top constraints had been safety and security, access to skills, and explainability. Adoption is already broad, with 86% of respondents reporting AI use, though the Bank cautions that the 2024 and 2026 populations differ.
This is our reasoning rather than the FCA's, but it follows from the shape of the programme. A synthetic dataset designed to reflect real-world scenarios is good for proving a method and poor at revealing the edge cases buried in a firm's own records. A cloud environment that is not connected to your core banking platform, case management system or CRM cannot show what happens when a model's output has to land in one of them. And a demo to a showcase audience is not a control test. Our AI solutions page lists six named solutions, from post-call follow-up automation to enterprise knowledge retrieval.
Data protection compounds it. In the same survey, 42% of respondents perceived UK data protection and privacy regulation as a high compliance burden. Bringing real customer data into any external environment is a decision for your DPO and your Consumer Duty and SM&CR owners, not a box ticked on an application form. If the build will eventually run on self-hosted models inside your perimeter, testing it somewhere else answers only part of the question.
Is the Supercharged Sandbox right for your AI build?
The Supercharged Sandbox suits a firm with a clear AI use case, a realistic testing plan for the cohort period, the in-house expertise to build independently, and a genuine need for compute, tooling or data it cannot easily get elsewhere. It suits a firm poorly when the real blocker is integration with live systems or internal sign-off, which a separate test environment cannot resolve.
The quickest self-test is to score your idea against the FCA's own five assessment criteria before anyone writes an application.
- Relevance and strategic fit. Is the work relevant to financial services, market infrastructure or compliance activity, and suited to development or validation in the Sandbox?
- Degree of innovation. Is the solution materially different from existing market practice, and does AI play a substantive role?
- Feasibility and readiness. Is the testing plan achievable within the cohort timeframe, and does the team have the capacity to take part?
- Potential impact. Are the outcomes for consumers, firms or markets clear, realistic and capable of being assessed, with a credible path to adoption or live testing afterwards?
- Need for support. Is there a clear need for this environment rather than general innovation support, and can the firm obtain the data the work requires?
The last two criteria deserve the most honest answers. A firm that cannot describe what happens after the showcase has no credible pathway, and a firm that could run the same experiment in its own Azure, AWS or Google Cloud tenancy struggles to show need. Answering both honestly is the kind of work a placement-led AI approach puts first: decide where the AI belongs, then decide where to test it.
The same sequencing runs through the DATS Placement Diagnostic, which produces a ranked roadmap of where AI belongs and where it does not. If you want a structured first pass before that, our mid-market AI readiness assessment walks through the self-run version.
What should a firm prepare before applying to a future cohort?
A firm should prepare a one-page use case, a testing plan that fits a cohort of five to six months, a named owner with time to take part, a view on which data it will use, and a stated route to production after the showcase. As of 14 November 2026 the FCA has not published dates for a further cohort, so this work can be done before any window opens.
The FCA's Sandbox page gives no date for a third cohort; cohort 2's showcase runs on 25 and 26 November and its sandbox access ends on 31 December 2026. Watch the Supercharged Sandbox page for any announcement, and treat the following as a pack you can assemble now.
- The use case in one paragraph. The problem, the customer or market outcome, and why AI is substantive to it rather than decorative.
- The testing plan. What you will measure, against what baseline, and what result would make you stop.
- The data position. Whether synthetic or curated data is enough, or whether you will bring your own data and what your data protection review says about that.
- The team. Who builds, given the FCA expects participants to build independently, and who owns the result.
- The pathway. What happens after the showcase: a production placement, a Regulatory Sandbox test, AI Live Testing, or a decision to stop.
Point five is where a sandbox project either becomes a production system or becomes a slide. Pilot to Production is one of the five stages of our DATS methodology, and the governance that has to be in place before a model touches customers is what our operating model work designs. If the pack shows the real constraint is internal, not technical, talk to us before you spend a cohort on it.
What is the best way to test an AI build in financial services in 2026?
The best route for testing an AI build in UK financial services in 2026 depends on what you need to learn. The Supercharged Sandbox is strongest for compute-heavy experimentation on curated or synthetic data, ending in a showcase to regulatory and industry stakeholders. Your own cloud environment is strongest for testing against real systems and data. The Regulatory Sandbox and AI Live Testing are for live conditions, which a pre-deployment sandbox cannot simulate.
Use four criteria to choose. First, what is unproven: the method, or the fit with your processes. Second, what data the test needs, and whether it can lawfully leave your perimeter. Third, what kind of regulatory contact helps: the Supercharged Sandbox offers expert support and a showcase audience, while the Regulatory Sandbox offers a case manager and individual guidance. Fourth, timing, since a cohort runs to the FCA's calendar rather than yours.
On those criteria, the Supercharged Sandbox wins when a smaller firm lacks GPU capacity or enterprise AI tooling, when the use case benefits from presenting to regulatory, industry and technology stakeholders at the showcase, or when synthetic data is genuinely sufficient to prove the method. A large bank with its own GPU estate and a model risk function may learn more, faster, by testing inside its own environment, because the test can run against its own systems, and can still take a mature proposition to AI Live Testing later. Neither route replaces the governance a production deployment needs, which our wider compliance articles cover in depth.
The macro picture argues for caring about the handover more than the experiment. McKinsey's State of AI put the share of organisations with AI in production at 33%. Getting an idea to work once is not the same as running it in a live process with an owner, which our enterprise AI consulting guide treats as the core of the job.
Frequently asked questions
Does a firm need FCA authorisation to join the Supercharged Sandbox?
No. The FCA says firms do not need to be FCA regulated to take part in the Supercharged Sandbox, provided their solution is relevant to UK financial services or markets. Firms based outside the UK can also apply on the same condition. Authorisation questions arise later, if a firm moves from experimentation to offering a regulated product or service to consumers.
That later step is where the Regulatory Sandbox and its restricted authorisation route become relevant, and where an AI operating model needs named owners for each control.
Does a Supercharged Sandbox place mean the FCA has approved the product?
No. The FCA states that taking part in the Supercharged Sandbox does not indicate FCA approval, endorsement or authorisation of a product or service. A firm should not describe its participation to customers or investors as regulatory sign-off. The Sandbox provides an environment and support for experimentation; it does not change the rules any resulting product must meet.
The FCA has also said it intends to rely on existing frameworks for AI rather than adding new AI-specific rules, so the obligations a sandbox project eventually meets are the ones already in force; in our reading that includes regimes such as Consumer Duty and SM&CR.
While no further Supercharged Sandbox cohort is announced, the permanent Digital Sandbox takes applications through its platform, and the FCA has also launched an Agentic Academy, a 10-week specialist AI programme for selected firms, delivered with the Centre for Finance, Technology and Entrepreneurship. For a view on which of your use cases deserves a test at all, read about our approach to placing AI.
Want a second opinion before you apply? See the DATS consulting system, review the enterprise AI solutions we build, read about Dilr.ai, or browse more AI compliance guides for regulated firms.
Test the use case that will reach production.
30-min scoping call · No deck · Confidential. We will tell you which AI use case deserves a test, and what it needs to reach production.
Written by the Dilr.ai engineering team, practitioners who ship enterprise AI in production. Follow us on LinkedIn for shipping notes, or subscribe via the RSS feed.