Anthropic to Support FCA's Second Supercharged Sandbox Cohort

The FCA lists five broad workstreams: safer agent-led payments and commerce, fraud and economic-crime detection, AI governance and accountability, access for vulnerable and underserved consumers, and compliance or business-process automation. Scottish Widows, Money Advice Trust and TrueLayer are among the participants.

The regulator says the cohort received 199 applications, up from 132 for the first group, a 51% increase. Existing support from NayaOne and NVIDIA will continue, providing digital-sandbox infrastructure and accelerated computing. The FCA also announced a 10-week Agentic Academy for selected firms, delivered with the Centre for Finance, Technology and Entrepreneurship.

The Banker independently confirms the second-cohort participation, the 21 organizations and the Claude tools being made available. FF News reports the same program details and use-case categories.

The announcement is about access and supervised experimentation. It does not say that the FCA has approved Claude for unrestricted deployment in regulated financial services, endorsed a participant's product, or relaxed existing obligations. The regulator describes the Supercharged Sandbox as support for firms in the discovery and experiment phase.

That distinction matters for agent-led payments and compliance automation, where a model could influence transactions, customer outcomes or regulatory decisions. A successful prototype would still need evidence that its data handling, model controls, accountability and consumer protections satisfy the rules that apply to the production use case.

The cohort can produce useful evidence if participants define measurable outcomes before testing. Fraud teams need false-positive and false-negative rates; payments teams need limits, approval paths and recovery procedures; governance teams need traceable prompts, tool calls, model versions and human decisions. Tests involving vulnerable consumers also require careful monitoring for exclusion or unequal outcomes.

The most valuable outputs will be published evaluation methods, failure cases and control patterns that other regulated firms can reproduce. Until those results appear, the July 22 announcement establishes the participants, tools and intended workstreams, not the effectiveness or regulatory readiness of any system they build.