evidoria

← Back to browse

Good practice Imported

FCA's Supercharged Sandbox — the UK Financial Regulator's AI Testing Cohorts With NVIDIA, AWS and Anthropic

United Kingdom · London · See the United Kingdom profile · See the London profile

Evidence: Observational / pre–post Top 74% 47/100 · Ask Evidence Copilot about this practice

The UK's Financial Conduct Authority runs 20-week 'Supercharged Sandbox' cohorts where selected firms stress-test AI systems on real infrastructure (NVIDIA, AWS, Anthropic's Claude) under regulatory observation; applications rose from 132 to 199 between its first two cohorts.

132
Cohort 1 applicants (Oct 2025–Jan 2026)
23
Cohort 1 firms selected (Oct 2025–Jan 2026)
20 of 23
Cohort 1 firms presenting at public showcase (28–29 Jan 2026)
199
Cohort 2 applicants (2026)
51 %
Rise in applicants, Cohort 2 vs Cohort 1
21
Cohort 2 firms selected (Jul–Dec 2026)
FCA's Supercharged Sandbox — the UK Financial Regulator's AI Testing Cohorts With NVIDIA, AWS and Anthropic

Details

Maturity
Scaling
Promoter
Financial Conduct Authority (FCA) — Innovation Services / AI Lab, with NayaOne, AWS, NVIDIA and Anthropic
Period
October 2025–ongoing (Cohort 1: Oct 2025–Jan 2026; Cohort 2: Jul–Dec 2026)
Keywords
financial regulation, AI governance, fintech, compliance technology

Context

The UK's Financial Conduct Authority (FCA) runs a 'Supercharged Sandbox' inside its Innovation Services / AI Lab: cohort-based programmes where selected firms build and stress-test AI systems on real infrastructure under regulatory observation, rather than in slide decks or lab conditions.

Objectives

The sandbox lets firms stress-test AI systems for use cases including agent-led payments, fraud and financial-crime detection, AI governance tooling, access to financial services for vulnerable and underserved consumers, and compliance automation.

Activities

Cohort 1 ran for 20 weeks from October 2025 to January 2026, with 23 firms selected out of 132 applicants (split between retail-innovation/consumer-insight and wholesale/compliance-innovation tracks) and technology support from AWS, NVIDIA and the Digital Sandbox platform NayaOne. 20 of the 23 firms presented their systems at a public showcase in London on 28–29 January 2026. Cohort 2 launched on 13 July 2026 and runs to 31 December 2026, with 21 firms selected from 199 applications (a 51% rise on Cohort 1), including Scottish Widows, Money Advice Trust and TrueLayer; Anthropic joined as a partner, giving participants access to Claude (including Claude Code and Claude Cowork). All experimentation used realistic but non-sensitive datasets with no live-system interaction.

Results

According to NayaOne's write-up, several Cohort 1 firms deliberately narrowed scope or rethought their architectures after evaluation findings.

Conclusions

Public sources document participation, process and infrastructure in detail but do not yet publish independently measured consumer-outcome or market-impact figures for either cohort; the programme's value should currently be read as a structured testing environment with growing take-up, not as a proven driver of measured public benefit.

Implementation
Indicative cost
Medium (€50k–€500k) — Not quantified in sources reviewed; delivered via in-kind/partnered infrastructure from AWS, NVIDIA, NayaOne and Anthropic rather than a stated cash budget.
Time to results
Medium (1–3 years) — Cohort 1: October 2025–January 2026 (20 weeks); Cohort 2: 13 July–31 December 2026.
Staffing & skills
FCA Innovation Services / AI Lab team (named: Matt Lowe, Innovation Lab manager), Technology partners: AWS, NVIDIA, NayaOne (Digital Sandbox platform), and Anthropic (Cohort 2)

Conditions for success

  • A structured cohort format with a fixed duration (20 weeks) and named selection tracks (retail-innovation/consumer-insight vs wholesale/compliance-innovation)
  • A realistic-but-non-sensitive data policy with no live-system interaction
  • Named technology infrastructure partners providing compute and platform support

Common failure modes

  • No independently measured consumer-outcome or market-impact figures are yet published for either cohort; current evidence is process/participation data rather than proven public benefit

Where it fits

Governance type
national financial regulator (FCA) with private technology partners
Scale
cohort-based (20–23 firms per round)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful