Moody’s has put a name to a risk building quietly inside the banking system: nearly every large bank is racing to install artificial intelligence, and nearly all of them are buying it from the same few companies.

The rating agency’s “Bank of the Future” analysis, published in late July, said AI will eventually cut costs and lift revenue across Wall Street and the City of London, with significant risk attached. Most financial firms now depend on a relatively small group of foundation AI model and cloud computing providers — what Moody’s calls a “systemic dependency” — and an outage at a single major provider could ripple across customers and entire sectors at once. Regulators, the agency expects, will sharpen their focus on operational resilience and third-party concentration as adoption deepens.

The distinction from ordinary technology risk matters. A bank typically runs several different suppliers across different systems. With generative AI, many institutions end up relying on the same underlying models, the same cloud infrastructure and the same handful of vendors. One failure then becomes everyone’s failure simultaneously, rather than one bank’s bad week.

The pricing problem

Moody’s also flagged what it termed vendor dependence risk — the prospect that a set of dominant model and infrastructure providers could, over time, control the price of AI services.

The report named OpenAI and Anthropic specifically, noting that both face investor pressure to reach profitability while still running losses — pressure Moody’s believes could eventually hand those vendors leverage over pricing terms with the institutions building on their models.

That is the sequence worth watching. A bank spends two years rebuilding fraud detection, credit decisioning and customer service around a particular model. Switching costs climb with every workflow moved over. Then the contract comes up for renewal, and the bank’s negotiating position is considerably weaker than it was at signing.

Moody’s added that the payoff may be thinner than banks expect: capturing it requires substantial investment, and with so many competitors chasing the same efficiencies, much of the gain gets competed away. Everyone spends, everyone gets faster, and the savings pass through to customers rather than to shareholders.

The deposit risk

One warning is specific to banking, and it should register with anyone who lived through March 2023. Moody’s said AI could make it far easier for depositors to move money into higher-yielding accounts, potentially shifting significant sums in a short period. That puts depositor trust and funding stability directly in scope.

Silicon Valley Bank collapsed in part because customers could move money out with a phone in their hands faster than the bank could raise liquidity. An AI assistant that continuously monitors rates across institutions and moves cash on its own instruction compresses that timeline further. Moody’s grouped this alongside heightened exposure to data privacy failures, cybersecurity gaps and fraud.

How deep adoption already runs

More than three-quarters of financial services firms in the United Kingdom already use AI, according to a Treasury select committee report. Lloyds Banking Group is the clearest large-scale commitment, with chief executive Charlie Nunn pursuing a £13 billion strategy that includes £2 billion in cost cuts, acknowledging the effect on staff and pledging continued reskilling alongside new hiring.

Moody’s also attached a figure to the displacement question: a one-in-five chance that AI can perform the work of a capable mid-level employee by 2030.

What banks can do about it

The agency did not leave the problem without remedies. Moody’s said banks and insurers can reduce their dependence by keeping control of their own data, applying their considerable experience negotiating technology contracts, using open-source models, and building partnerships rather than single-vendor relationships.

That first item is the one most within reach. A bank’s proprietary data — its lending history, its customer behavior, its fraud patterns — is the asset the model providers cannot replicate. Institutions that keep that data under their own control and portable between systems retain the ability to walk. Those that let it settle inside a vendor’s platform are the ones who will find the renewal conversation unpleasant.

The open-source option has also become materially more credible in the past few months, with capable models now available under permissive licenses that run on hardware a bank already owns. For a mid-sized institution weighing a first AI deployment, that is worth evaluating before signing a long-term commitment to any single provider.

The broader point Moody’s is making is not that banks should slow down. It is that concentration risk is the thing regulators eventually price, and that the industry is building it right now, in plain view, one vendor contract at a time.

JBizNews Desk | New York

© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.