Every prior update to this pilot added a handful of institutions, usually within countries already tracked. This one is different in kind, not just degree: 24 new institutions and 28 new vendor relationships, spanning 17 countries the pilot had never touched, bringing the total to 93 institutions, 120 documented vendor relationships, across 27 countries.
Where the pilot is now: five continents
New entries this round: Spain, the UAE, Brazil, Italy, Canada, South Africa, Ireland, Denmark, Poland, Austria, Indonesia, Thailand, Taiwan, Chile, Colombia, Argentina, and Nigeria. Latin America, Sub-Saharan Africa, and the Middle East all enter the sample for the first time in this single pass β regions the pilot had no visibility into as recently as this morning.
The headline finding: Santander runs on four foundation models at once
Banco Santander enters with the sample's first four-way confirmed relationship: Microsoft (via Microsoft 365 Copilot), OpenAI (ChatGPT), Anthropic (Claude), and Google (Gemini), all named explicitly by Santander's own Chief Data & AI Officer, Ricardo MartΓn ManjΓ³n, in an official company story announcing AI access for all 185,000 employees. Read against the family-not-vendor-name methodology, Santander is the cleanest real-world test case yet of that principle: a genuinely diversified vendor strategy, all four families independently confirmed in one official source rather than stitched together from separate leads.
A pattern worth naming honestly: this batch skews heavily Microsoft
Of the 28 new relationships in this update, 25 are Microsoft (Azure OpenAI), mostly via Microsoft 365 Copilot or GitHub Copilot. That is not a claim that Microsoft objectively dominates AI adoption in Brazilian, Nigerian, or Indonesian banking β it is a visible artifact of where the evidence is easiest to find. Microsoft's own "Customer Stories" program publishes a steady stream of dated, quoted, official case studies for exactly the banks and insurers this pilot tracks; OpenAI, Anthropic, and Google have nothing at the same publication scale for regional retail banks outside their most visible flagship accounts. That's a sourcing-availability bias worth stating plainly rather than letting the raw numbers imply something about the underlying market that the search methodology can't actually support. Intesa Sanpaolo (Italy, confirmed on Gemini via an official Google Cloud case study) is this round's only non-Microsoft find β worth noting precisely because it's the exception.
One institution held to a lower tier on purpose
CTBC Bank (Taiwan) uses Microsoft Copilot internally with reported gains β 150+ use cases, up to 70% efficiency improvement β but the only source found is trade press (The Asian Banker), not an official Microsoft or CTBC case study. It's marked Probable, not Confirmed, consistent with how every prior update has handled thinner sourcing rather than rounding it up because the underlying story is plausible.
What's next
This expansion was found in a single research pass, not spread across weeks β which means the same pass almost certainly missed real, sourced relationships in the countries it did cover, especially outside banking (insurers and asset managers remain thin in every country added this round except Canada and France). See the full data table for every row and its source, and the methodology for what counts as evidence at each confidence level.