FCA Adds Anthropic Tools to Second Supercharged Sandbox Cohort
Twenty-one organisations will test agent-led payments, fraud detection, governance, inclusion and compliance workflows using Claude tools in a controlled FCA environment.
The UK Financial Conduct Authority announced on July 22 that Anthropic will support the second group in its Supercharged Sandbox, a controlled environment for firms testing advanced artificial intelligence. The FCA's official announcement says participants will receive access to Claude, Claude Code and Claude Cowork to accelerate development work.
The cohort brings a frontier-model provider directly into a regulator-operated experimentation programme. That gives participating firms a place to develop agentic finance tools while the FCA observes how the systems behave within a bounded environment rather than after a broad commercial rollout.
What the cohort will test
The FCA identified five areas for the second group. Participants will explore safer agent-led payments and commerce; improved detection of fraud and economic crime; stronger AI governance and accountability; wider access to financial services for vulnerable and underserved consumers; and automation for compliance and business operations.
Those categories span both customer-facing and internal uses. Payments and access work can affect how people interact with financial services, while fraud, governance and compliance projects focus on controls around those interactions. The breadth of the list also means the cohort is testing several kinds of risk rather than treating all financial AI as one use case.
The second group includes 21 organisations. The FCA said applications increased by 51 percent compared with the first group, with 199 applications for this cohort versus 132 in the first iteration. Existing support from NayaOne and NVIDIA continues, adding sandbox infrastructure and accelerated-computing access alongside Anthropic's model and agent tools.
The regulator also launched an Agentic Academy, described as a separate ten-week specialist AI programme for selected firms. That adds a training component to the controlled technical work and may help participants develop a shared language for governance, testing and deployment decisions.
Why it matters
Regulatory sandboxes are valuable when they turn general AI principles into observable workflows. The FCA's chosen test areas put accountability, fraud detection and inclusion beside faster development and automation. That framing matters because financial-sector adoption cannot be judged only by whether an agent completes a task; reviewers also need to understand who can inspect, interrupt and challenge its actions.
This announcement confirms the tools, participants and test themes, but it does not establish that any of the proposed systems are ready for customers. A sandbox cohort is an evaluation setting, not a general market launch or proof of safe performance. The useful next evidence will be published test methods, governance findings, failure cases and any changes firms make before broader deployment.
Status
Confirmed. The FCA officially announced Anthropic's role, the size of the second cohort and the five test areas. Outcomes remain unproven until the programme reports results.
Sources
Update note: Last reviewed 2026-07-29. We will revise this post when the FCA publishes cohort findings, evaluation methods or deployment decisions.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.