Every previous update to this pilot either launched it or took it somewhere new — a new country, a new model family. This one is different on purpose: 9 new institutions and 12 new vendor relationships, all within the 10 countries already tracked, bringing the pilot to 69 institutions and 92 documented vendor relationships.
How we found the gaps
Rather than searching institution-by-institution, we cross-tabulated the existing sample by country and category. That surfaced some genuinely odd holes: Switzerland — home to two of the world's largest wealth managers — had zero tracked banks. The Netherlands had insurers and banks, but no asset manager. Hong Kong had an insurer, but no bank or asset manager. Each of those gaps became a specific research target instead of a generic "find more institutions" search.
UBS closes the biggest gap
UBS enters the sample via its own case study: "UBS Red," a set of domain-specific AI assistants built on Azure AI Search and Azure OpenAI Service, rolled out to 30,000 employees across five divisions by December 2024, with roughly 60,000 investment-advice documents digitized. For a bank of UBS's size to have had zero presence in this index until now says less about UBS and more about how much of this pilot's growth still depends on which searches happen to surface a source — the same limitation every update to this index has been honest about.
Allianz joins a small group of three-family institutions
Allianz SE — already tracked for its Anthropic partnership — gains a second confirmed relationship this round: OpenAI's GPT-4o and DALL-E, delivered through AllianzGPT on Microsoft's Azure Cloud. That gives Allianz three simultaneously confirmed model families (Anthropic, OpenAI, Microsoft) — joining, not leading, a short list that already included Raymond James Financial, Citizens Financial Group, and MUFG from earlier updates.
It's worth reading that number the way the methodology insists on: three vendors on paper doesn't mean three independent points of failure. If Allianz's OpenAI access and its Microsoft access both ultimately run through Azure infrastructure, the real diversification is closer to two independent paths than three. Nomura Holdings, new to the sample this round, is the sharper contrast: it runs OpenAI's Deep Research for investment research and Anthropic's Claude (via AWS Bedrock, not Azure) for its internal "Core AI" compliance platform — two genuinely separate delivery paths for two different jobs.
AXA also picks up a second family this round — Mistral AI, integrated into its "AXA Secure GPT" platform — continuing the European-sovereignty pattern first noted when Mistral joined as a fifth tracked family: reduce dependence on US-based model vendors specifically for European institutions.
What didn't make the cut
Several searches this round turned up nothing citable, and we're noting the misses the same way we always have. Great Eastern Holdings (Singapore), QBE and AMP (Australia), DWS and Amundi (Germany and France's largest asset managers), and most Hong Kong banks beyond CSOP Asset Management — Hang Seng Bank, Bank of China (Hong Kong) — had no sourced foundation-model relationship we could find. Julius Baer has an official Microsoft case study, but it documents a developer cloud platform migration with no generative-AI product named, so it stayed out. And GIC, Singapore's sovereign wealth fund, led a $30B investment round in Anthropic this year — but being a shareholder in a model lab isn't the same as being a documented user of its models, so that relationship stays out of this index by design.
See the full data table for every row and its source, and the methodology for what counts as evidence at each confidence level.