Anthropic joins the UK FCA sandbox, with Claude directly plugging into AI payment and financial regulatory experiments

The UK Financial Conduct Authority (FCA) officially launched the second edition of “Supercharged Sandbox.” Anthropic will bring its Claude model into the sandbox, and 21 organizations will test AI payments, fraud detection, and compliance automation, with application volume up 51% compared with the first cohort.
(Background: The UK FCA warned about a “race to arm” in financial AI—pushing regulation to rein in large models like ChatGPT, Claude, Gemini, and others.)
(Additional context: Anthropic launched “Claude for Small Business,” targeting AI automation for small and medium-sized businesses—helping you speed up invoicing, calculate payroll..)

Table of contents

Toggle

• Claude model enters the sandbox
• Dual track in parallel: sandbox + real-world practice
• Taiwan’s financial AI regulation comparison

The UK Financial Conduct Authority (FCA) announced on July 22 that the second edition of “Supercharged Sandbox” is officially underway. AI startup Anthropic will bring its Claude model into the sandbox for participating organizations to develop financial AI applications.

This continues the FCA’s efforts in cooperation with NVIDIA and NayaOne—directly bringing AI foundation models into the financial regulatory testing environment. The second sandbox cohort includes 21 organizations, including Scottish Widows, Money Advice Trust, and payment infrastructure company TrueLayer.

Claude model enters the sandbox

Participating organizations will be granted access to the Claude model, including tools such as Claude Code and Claude Cowork. The development directions include:

• AI agent-led payment systems
• Fraud detection and compliance automation
• AI governance and oversight mechanisms

The FCA said this cohort received 199 applications, up 51% from the first edition, reflecting that demand from financial institutions to test AI systems in a regulated environment continues to rise.

Dual track in parallel: sandbox + real-world practice

Supercharged Sandbox is positioned in the “exploration and experimentation” stage of AI development. The FCA also runs another “AI Live Testing” program to support institutions that are ready to deploy AI into real markets.

In addition, the FCA and the “Centre for Finance, Technology and Entrepreneurship” jointly organized a 10-week Agentic Academy to help financial institutions develop and deploy AI agents.

Taiwan’s financial AI regulation comparison

Taiwan’s Financial Supervisory Commission is also promoting AI financial applications, but it still mainly relies on pilot tests conducted by individual banks. The FCA’s sandbox approach embeds the regulatory framework into the experimentation process earlier, allowing institutions to test new products in a controlled environment before rolling them out to the market—an approach Taiwan’s regulators could learn from.