TL;DR: Ant International has released version 2.0 of its Falcon Time-Series Transformer model with six large banks signed up, HSBC, Standard Chartered and Barclays among them, alongside Citi and Deutsche Bank. Kelvin Li, who runs platform technology at the Singapore-based fintech, claims precise forecasting takes upwards of 60% off what banks spend hedging currency exposure and allocating against it.

The specialisation argument is the interesting part. Li contends that general-purpose large models have yet to achieve anything like a universal breakthrough in finance, and that a model built for financial scenarios holds an edge over them. That is a narrower and more testable claim than most banking AI announcements make, and it points at where the sector’s spending is actually going: not chatbots, but time-series forecasting against liquidity risk, where a measurable error rate exists and hedging costs provide an unambiguous scoreboard.

For three UK-listed banks, the supply chain deserves a second look. Ant International is the overseas arm of Ant Group, the Chinese fintech founded by Jack Ma, and closed a $1.2bn equity raise last month. A Chinese-built model sitting inside live treasury operations at British institutions is precisely the arrangement the sovereignty debate keeps circling — and it has arrived through commercial procurement rather than policy, while Westminster is still arguing over what sovereign capability means.

That is not an argument against the deployment. Treasury forecasting is a well-bounded task with outputs traders can check, which puts it a long way from the open-ended agentic deployments drawing regulatory attention. But it does illustrate how quickly the question moves from theory to practice. Resultsense reported this week that Whitehall is assessing what losing access to frontier models would cost, and separately that Red Hat is urging ministers to define sovereignty by operational control rather than by a supplier’s nationality. Six banks have already made a working answer to that question.

Looking forward: Whether the 60% figure holds outside Ant’s own testing is the thing to watch, and the banks are unlikely to publish. The clearer near-term signal is competitive: if specialist financial models beat general frontier models on forecasting work, the procurement logic across UK treasury functions shifts away from a single flagship vendor.