Rathi's AI pivot: the FCA bets on competition as the new prudential frontier

Nikhil Rathi's speech at techUK on 24 June marks the most explicit shift yet in how the FCA intends to supervise AI in regulated firms. More than 80% of financial services firms are already adopting AI, according to studies cited by the FCA chief executive, and the question now, he said, is one of scale (FCA). The framing matters: Rathi is no longer treating AI as an emerging risk to be mapped, but as an active force reshaping market structure faster than the rulebook governing it.

Two scaling vectors anchor the speech. The first is agentic systems, which Rathi described as moving beyond summarisation and detection toward coordination and transaction, including liquidity management and trading workflows in wholesale markets (FCA). The second is tokenisation, where the FCA approved Baillie Gifford alongside Bank of New York Mellon on Monday to launch the UK's first natively tokenised authorised fund (FCA). The combination is significant: programmable infrastructure plus autonomous agents implies a market structure where execution, settlement and decision-making compress into a single technical stack. The regulatory question of who is accountable when an agent transacts becomes the central conduct question of the next cycle.

The more consequential signal sits in Rathi's framing of competition and resilience as linked concerns. As AI reshapes markets and increases interconnection, he said, understanding how competition is evolving, and where that may impact resilience, will become more important than ever (FCA). Read alongside the FCA's parallel competition action this week, where 11 commodity futures day traders offered a £1m ex gratia payment to the Crisis and Resilience Fund to settle concerns about coordinated trading and information sharing (FCA), the direction is clear. The FCA is positioning competition law as a live supervisory tool, not a dormant one, and intends to apply it to the concentration risks AI adoption is already creating in model providers, cloud infrastructure and data vendors.

For senior leaders, the practical implications are sharper than the speech's tone suggests. Boards that have framed AI governance primarily around model risk and consumer outcomes now need to add a third dimension: dependency on a small number of upstream providers, and the resilience consequences when those providers move in concert. Rathi was explicit that investors will be wary to delegate important decisions to systems they do not understand, and that accountability for regulated activities and outcomes must remain clear (FCA). That sentence will be quoted back at chief executives in supervisory letters. It means human oversight design, audit trails for agent decisions, and contractual clarity with AI vendors are no longer engineering concerns, they are board-level accountabilities.

The tokenisation track adds urgency. With banks already piloting tokenised deposits to reduce friction and fraud in processes like home buying (FCA), and the Baillie Gifford approval setting a live precedent, asset managers and custodians without a credible tokenisation roadmap will find themselves answering harder questions from distribution partners within twelve months.

The implication for executives is straightforward. The FCA has told the market it will measure AI adoption through the lens of competition and concentration, not just conduct. Firms that cannot articulate their upstream dependencies, and the resilience plan when those dependencies fail simultaneously, should expect that gap to surface in the next supervisory cycle.