Anthropic, the AI company behind the Claude model family, has signed on to the UK Financial Conduct Authority's second Supercharged Sandbox cohort. The program, which launched recently, lets financial firms test AI applications under the regulator's supervision. Anthropic will supply its Claude models to participating companies, giving them access to large language model technology within a controlled environment.
What the sandbox does
The FCA's sandbox is not a physical space but a regulatory framework. Companies get a time-limited, restricted authorization to try out new products or services without immediately facing the full weight of financial rules. The first Supercharged Sandbox cohort ran in 2023 and focused on general fintech innovation. This second cohort zeroes in on AI — specifically, how banks, insurers, and other financial firms can deploy generative AI safely.
Participants must still meet certain consumer protection and data privacy standards. But the sandbox gives them room to experiment, fail, and iterate without triggering enforcement actions. The FCA says it wants to understand where AI creates real value and where it introduces new risks.
Anthropic's role in the cohort
Anthropic is not a financial services company, but it provides the underlying AI technology. Through the sandbox, the company will make its Claude models available to selected fintech firms. That means those firms can build AI-powered tools — for customer service, fraud detection, compliance, or trading — using Claude's language capabilities, while the FCA watches how the models behave in a financial context.
Anthropic has previously emphasized safety and interpretability in its models. The company's participation suggests it sees financial services as a key sector for AI deployment, but one that requires careful oversight. The FCA, for its part, gets a direct look at how a frontier AI model performs under real-world regulatory constraints.
Financial services are heavily regulated, and AI adoption has been cautious. Banks worry about bias in lending algorithms, hallucinations in customer-facing chatbots, and opaque decision-making that regulators can't audit. The sandbox gives both the FCA and participating firms a chance to test these issues before they scale.
For smaller fintech startups, access to a model like Claude — normally priced for enterprise customers — could lower the barrier to entry. For the FCA, the program generates data on what works and what doesn't, which could shape future AI rules in the UK.
The second Supercharged Sandbox cohort is now open. Participating companies will run their tests over the coming months. The FCA has not set a specific end date for the cohort, but it typically runs for six to twelve months. After that, the regulator will publish findings and possibly update its guidance on AI in financial services.
Anthropic's involvement is limited to providing the models; the company is not a sandbox participant itself. How the FCA uses the results — and whether it tightens or loosens AI rules based on what it learns — remains an open question.
