Anthropic is putting Claude inside one of the United Kingdom’s most closely watched financial technology testing programmes.
The AI company will support the second cohort of the Financial Conduct Authority’s Supercharged Sandbox, giving participating organisations access to Claude, Claude Code and Claude Cowork. The tools will sit alongside infrastructure already supplied by NayaOne and Nvidia.
This is not a broad experiment involving a few generic banking chatbots. The firms involved will test ideas around fraud detection, automated compliance, financial inclusion and payments carried out by AI agents. Those are areas where mistakes can become expensive very quickly.
Claude Joins the FCA Supercharged Sandbox
The FCA created the Supercharged Sandbox as a controlled environment where companies can build and test advanced AI products without immediately exposing real customers or financial markets to unfinished technology.
Participants receive access to GPU-enabled computing infrastructure, enterprise AI tools, synthetic datasets and support from specialists in regulation, financial services and artificial intelligence. The idea is to move companies beyond PowerPoint concepts and into actual testing.
Anthropic adds another layer to that setup. Claude Code can help development teams write, review and debug software, while Claude Cowork is designed to support more general workplace and operational tasks. For firms with small technical teams, that could cut down some of the work required to get an early product running.
FCA Selects 21 Organisations From 199 Applications
The FCA selected 21 organisations for the second sandbox cohort after receiving 199 applications. That represented a 51% increase from the number submitted for the first programme.
The group includes Scottish Widows, Money Advice Trust, TrueLayer, Sardine AI, Trustie Labs, Deepflow, FSCom and Ubyx, which is participating alongside Amazon. Other selected organisations include Aegis Trace, calQrisk, Condukt, GAI Labs, IntelXview, Kaption, Merx Digital Solutions, Relace, RMI Agentic, Welleness and Zquas.
There is a mix here. Established financial services companies are working beside regtech providers, fraud prevention businesses and smaller AI developers. That should give the FCA a wider view of how the technology behaves across different parts of the market, not just inside large banks.
Fraud Detection and Agent-Led Payments Take Priority
Several participants will work on systems intended to detect fraud and economic crime more effectively. Financial criminals already use automation, synthetic identities and rapidly changing payment methods. Manual review teams struggle to keep pace, especially when suspicious activity hides inside enormous volumes of legitimate transactions.
AI could help spot patterns earlier. It could also generate a flood of false alarms or make decisions that investigators cannot properly explain. The sandbox gives firms space to work through those problems before attempting a wider rollout.
Agent-led payments will be another major area of testing. These systems allow an AI agent to initiate or manage a transaction on behalf of a user. It sounds convenient until questions appear. Who authorised the purchase? What happens when the agent misunderstands an instruction? Who carries responsibility when something goes wrong?
The participating firms are expected to test ways of making this type of commerce safer, rather than treating autonomous payments as another feature that can be launched first and fixed later.
Compliance Automation Could Be the Immediate Opportunity
The less glamorous use cases may prove useful first.
Financial institutions spend large amounts of time reviewing documents, checking customer information, monitoring transactions and preparing regulatory reports. Much of that work follows repeatable processes, but it still requires careful handling because one missed detail can create legal or financial consequences.
Claude could help companies organise documents, analyse internal information, generate code and automate parts of those workflows. Human oversight will still matter. No serious regulator is likely to accept “the AI handled it” as an explanation for a failed compliance process.
What the sandbox can reveal is where automation genuinely saves time and where it simply moves the risk somewhere less visible.
Financial Inclusion Is Also Part of the Test
The programme is not focused only on operational efficiency.
Some participants will explore how AI could widen access to financial services for vulnerable and underserved consumers. That may include improving financial guidance, identifying customers who need additional support or making complicated financial information easier to understand.
The opportunity is obvious. So is the danger.
A model trained on incomplete or biased information could provide weaker recommendations to the same people the technology is meant to help. Testing these systems in a controlled setting gives the FCA and participating firms a chance to examine how decisions are produced, who may be excluded and whether consumers can challenge an automated outcome.
The FCA Is Watching How Fintech AI Develops
The FCA is not only giving businesses access to expensive computing tools. It is also collecting evidence.
By observing how companies build AI products, the regulator can see where existing rules work, where uncertainty slows development and where new risks may require a different response. That matters because regulation often moves after technology has already entered the market.
The Supercharged Sandbox offers an earlier look.
Still, participation does not mean the FCA has approved, endorsed or authorised any product being developed through the programme. The regulator states this clearly. Companies will need to meet the usual legal and regulatory requirements before taking their services into the live market.
Testing Will Continue Until the End of 2026
The second cohort officially launched on July 13, 2026, and will continue until December 31, 2026.
Participants are scheduled to present their work during a demonstration day on November 26. The event will bring together financial services companies, regulators and technology specialists to examine what the firms built and what they learned.
Not every project will become a commercial product. That is partly the point. A useful sandbox should expose weak ideas before customers do.
For Anthropic, the programme puts Claude closer to real financial services development. For the FCA, it creates a clearer view of how agentic AI, compliance automation and fraud detection may work outside a controlled demo. The interesting part will come when those prototypes meet actual regulation, actual money and actual customers.
Sources
- FinTech Futures – Anthropic to support FCA’s Supercharged Sandbox
- Financial Conduct Authority – Supercharged Sandbox
- Financial Conduct Authority – FCA allows firms to experiment with AI alongside Nvidia
- Anthropic – Build with Claude