From joining the FCA’s first AI Sandbox to mentoring its next.

Today, the Financial Conduct Authority launched the second AI Supercharged Sandbox cohort. We know the excitement well; 10 months ago, we were building in the very first cohort.

We didn’t have much time to talk about it at the time - we were still in stealth, heads down, building and testing capabilities. It was a memorable time. On the exact day we were told BEYLA had been chosen to join the cohort, one of our co-founders underwent a scan that revealed an aggressive form of cancer. So it’s fair to say the last 10 months have been life-changing.

Today feels like the right moment to talk about this. Most importantly, Catherine was able to get early treatment and, after a number of procedures, is now healthy, and driving BEYLA’s brand and marketing. BEYLA is in great shape too: the team has grown, and we have an active beta programme with test clients using BEYLA to proactively run their business, surface insights, and make better decisions. We have signed a multi-year agreement with a global card scheme to allow BEYLA to continue developing and launching the agentic movement of money, currently in private development. And we have forged exciting partnerships that will position BEYLA as a trusted leader in sovereign AI.

Many of these achievements may not have been possible had we not had the opportunity to build and test BEYLA in the Supercharged Sandbox. We see it as a pivotal part of our story, and it’s why, when the FCA asked if I’d be open to mentoring firms in the next cohort, I didn’t hesitate in saying yes - you have my full support.

What the AI Supercharged Sandbox is

The AI Supercharged Sandbox is the FCA’s environment for letting financial services firms experiment with AI safely, with the regulator watching. Launched in June 2025 in partnership with NayaOne, AWS and NVIDIA, it gives firms access to serious accelerated computing power and high-quality datasets inside a secure, supervised space - room to test early ideas without putting real customers or market integrity at risk. For this second cohort, Anthropic and Google Cloud have joined the programme.

That combination matters more than it sounds. Advanced AI needs real compute and real data to be tested properly, and most early-stage firms have neither at the scale required. The Sandbox closes that gap. It hands you the infrastructure to build something ambitious, and the guardrails to build it responsibly. You get to move quickly without moving recklessly.

The inaugural cohort was very competitive: 23 firms selected from 132 applications, building and stress-testing from October until the Supercharged Showcase in late January 2026, a full-day event held at the FCA’s office in East London.

BEYLA was chosen as one of those 23.

Why we brought BEYLA into the room

Our thesis is a bold one. Finance should act for you, not wait to be asked. That is a big claim to make about people’s money.

And big claims about money deserve to be tested. Not in a pitch deck. In an environment built for scrutiny.

So that is what we did. We used the Sandbox to pressure-test the foundations of an AI-native business and finance operating system: proactive intelligence powered by the Hive, delivered through Digital Humans, with domain expertise, context and memory. BEYLA surfaces insights and manages administrative tasks, while the member stays in control of every decision.

The point was never to prove the AI could be clever. The point was to prove it could be trusted to be accurate, repeatable and responsible.

What we were really testing: trust

Because the hard part of AI in finance was never the intelligence. It’s getting access to the right data, at the right time, with the right context, and, most importantly, building a trust layer that ensures the outcomes are clear, repeatable and responsible.

Can every recommendation be explained? Can every action be traced back to real data? Does the human stay in control when it matters most? These are the questions a regulator asks. They are also exactly the questions a small business owner should ask. Testing against them made BEYLA sharper, safer and more honest.

A model that gives a confident answer is easy. A model that can show its working, cite the data it relied on, and hand the final decision back to a human using language that matches their level of experience and domain knowledge - that is harder. And it is the only version worth building when the outcome is someone’s payroll, their next hire, or whether they take a risk that could make or break the year.

Everything we have built is consent-based, explainable and auditable. The regulated services that move money - payments, cards, foreign exchange - remain in private development ahead of licensing.

Building the right way, not just the fast way

There is a version of AI in finance that rushes to market, promises the world, and quietly hopes the accuracy catches up later. We have never wanted to build that.

Proactive intelligence across a whole business only earns its place if people can rely on it. So we anchored our work in the Sandbox on the things that are easy to skip and hard to retrofit: explainability, auditability and human control. Get those right early and everything you build on top inherits them. Bolt them on late and you are forever apologising for the gaps.

Testing in a regulated environment forced that discipline on us from the start. We are grateful it did.

From participant to mentor

10 months on, the programme has grown. Applications to the second cohort rose 51%, to nearly 200, with 21 firms selected (FCA). Interest in building AI the responsible way is accelerating, not cooling.

And this time, I’m on the other side of the table - mentoring the next group of builders coming through, and cheering them on. I know some of the questions they will wrestle with, because we wrestled with them 10 months ago. You test, you learn, you reflect, you help the next ones test better. Building technology is often about iteration, and there’s something fitting in that.

Why this matters if you run a small business

None of this is abstract. For the owner of a small business, it’s the difference between AI that has been stress-tested with a regulator watching, and AI that has been rushed to market to catch a wave.

When the technology is making sense of your cash flow, your risk and your next decision, you deserve to know it was built to a standard - not shipped on a hunch by a team that has never built a career in finance. Regulatory scrutiny is not a brake on innovation. Handled well, it is a mark of it.

We chose what some might consider the harder path. More demanding, slightly slower, and in full view of the regulator - but it is the only honest one when the stakes are someone’s livelihood.

We joined the FCA’s first AI Supercharged Sandbox not for the badge, but because building intelligence that acts on your finances is a responsibility before it is a product. That conviction is why we were in the room 10 months ago, and why we’re helping others into it now.

Frequently asked questions