The Correlation Index publishes its first version today: a methodological pilot that documents, with verifiable public sources, the relationships between ten US financial institutions and the AI foundation model vendors they use.
What the pilot shows
Across this initial sample — JPMorgan Chase, Goldman Sachs, Citigroup, Bank of America, Wells Fargo, Morgan Stanley, AIG, Visa, BlackRock, and BNY Mellon — 100% of tracked institutions have a confirmed relationship with OpenAI and/or Anthropic. Two vendors, on their own, cover the entire sample.
The breakdown by model family:
- OpenAI (GPT) — 80% of institutions
- Anthropic (Claude) — 60%
- Google (Gemini) — 20%
- Microsoft (Azure OpenAI) — 10%
This is exactly the signal the index is designed to surface: a diversity of storefronts — ten distinct, seemingly competing institutions — sitting on a very small number of real model vendors.
Why it matters
As critical financial decisions (credit, underwriting, trading, fraud detection) get delegated to AI agents, this concentration becomes a potential systemic risk: institutions that look diversified can fail in a correlated way if their agents rely on the same underlying model.
The limits, honestly
Ten institutions is a pilot — not definitive statistical proof. Each relationship is classified by confidence level (Confirmed, Probable, Unverified) and every row of the data table links to its source. See the full methodology for the details on criteria and limitations.
What's next: expanding the sample toward identified targets (U.S. Bancorp, PNC, Truist, Capital One, and others), and launching the empirical correlation test between models on risk-decision scenarios.