Anthropic joins the FCA sandbox: AI supervision moves from policy to practice

The FCA has confirmed that Anthropic will support the second cohort of its Supercharged Sandbox, providing participants with access to Claude, Claude Code and Claude Cowork to accelerate development (FCA). Twenty-one organisations have been selected, including Scottish Widows, Money Advice Trust and TrueLayer, from a field that grew 51% on the first cohort, with 199 applications received against 132 previously (FCA). The regulator has also launched an Agentic Academy, a 10-week programme delivered with the Centre for Finance, Technology and Entrepreneurship (FCA).

Supervision by participation

The more interesting story is not the technology partner but the supervisory posture. The FCA has told the Prime Minister it will avoid additional regulations for AI by relying on existing frameworks (FCA). That is a significant choice. Rather than build a new AI rulebook, the regulator is embedding itself alongside firms during discovery and experimentation, shaping norms in the room rather than through consultation papers. The use cases named, safer agent-led payments and commerce, fraud and economic crime detection, AI governance and accountability, access for vulnerable consumers, and compliance automation, are precisely the areas where supervisory expectations would otherwise be written after the fact (FCA).

For boards, this changes the calculus on AI programmes. Firms inside the sandbox will benefit from privileged sightlines into how the FCA thinks about agentic systems, model risk and consumer outcomes, delivered through access to NVIDIA compute, NayaOne infrastructure and now frontier model providers (FCA). Firms outside will be inferring those expectations from public statements. Jessica Rusu, chief data, intelligence and information officer at the FCA, framed this as central to the commitment to economic growth and to maintaining the UK's position at the forefront of responsible AI adoption (FCA). The growth mandate the FCA has been handed is now being spent, visibly, on AI infrastructure.

The stakeholder question

The inclusion of a major life insurer alongside a payments infrastructure firm and a debt advice charity is deliberate. It tells procurement teams, general counsel and chief risk officers that AI governance is not going to be a sector-specific exercise. Model deployment for a vulnerable customer triage tool at Money Advice Trust and for institutional payments at TrueLayer will be examined against the same underlying principles (FCA). That has knock-on effects for third-party risk assessments, for how firms document model provenance, and for how board committees receive assurance on systems they did not build.

It also raises a competitive concentration question. The sandbox now runs on infrastructure from NayaOne and NVIDIA, with a frontier model from Anthropic (FCA). Firms building serious AI capability will find themselves aligned, whether by choice or by default, with the same small group of providers the regulator has effectively endorsed through participation. Procurement decisions taken in the next twelve months will be difficult to reverse.

The implication

AI supervision in UK financial services is being written in real time, in a room most firms are not in. Leaders who want influence over the rules their competitors will operate under need to be applying for the third cohort, not waiting for the policy statement.