The Correlation Index β global pilot
A handful of foundation models, a diversity of storefronts.
As critical financial decisions β credit, underwriting, trading, fraud β get delegated to AI agents, institutions that appear to compete with each other can fail in a correlated way when the same shock hits. We measure this concentration with public evidence and an empirical correlation test between models.
0%
of tracked institutions have a confirmed relationship with OpenAI and/or Anthropic, across a pilot sample of 204 institutions worldwide β outside the US, Microsoft and Google show a noticeably stronger foothold.
Live concentration map
204 institutions, a handful of hubs.
Each line is a tracked institution. Each hub is a foundation-model family, sized by how much of the pilot sample it covers. Hover to tilt the map.
0%
Microsoft (Azure OpenAI)
0%
Anthropic (Claude)
0%
OpenAI (GPT)
0%
Google (Gemini)
0%
Mistral AI
An institution can appear in several families at once β see what the index does not measure.
Mechanism 1
Dependency mapping
Public institutionβvendor relationships, classified by model family rather than commercial name.
See the data explorer β βMechanism 2
Empirical correlation test
Same risk-decision scenarios submitted to several foundation models, observed agreement rate compared to chance.
See the results β βProduct
Beta access β free, open now.
Build your own perimeter of institutions, see how concentrated its AI-vendor dependencies are, and test what one vendor failure would reach. Free account, no invite.