Banks Begin Pricing Generative AI Risk as Euro-Area Study Links Adoption to Shifting Default Probabilities

As generative artificial intelligence surges from novelty to boardroom priority, one problem keeps resurfacing for lenders: how to tell, in real time, whether AI is a growth engine that reduces risk—or a disruptive force that raises it. New research focusing on euro-area private companies suggests banks are already making that call, and that their interpretation has begun to show up in one of the most consequential numbers in credit markets: the probability of default.

The study examines lender-specific probability-of-default (PD) estimates—banks’ internal assessments of how likely a borrower is to fail—for private firms across four large euro-area economies. Because these companies are not continuously priced by public equity markets, bank risk models and credit committees often serve as a primary “translation layer” between technological change and financial conditions. That makes banks a revealing place to look for early signals of how generative AI is being priced.

A measurable shift after ChatGPT

The central finding is that, following the sharp rise in public attention to AI after the release of ChatGPT, banks reduced their assessed default risk for firms operating in industries with high exposure to AI. On average, PD estimates for these borrowers fell by about 5%, a meaningful change in a metric that can directly influence loan pricing, credit limits, collateral demands, and covenant tightness.

The timing matters. Generative AI’s mainstream breakthrough created a sudden influx of information—media coverage, corporate announcements, investor narratives and technology roadmaps—well before most firms could report audited, AI-related performance improvements in financial statements. In that information gap, lenders appear to have updated their beliefs: AI-exposed industries were treated more as a tailwind than a headwind.

Researchers frame the broader issue as a familiar one in economic history. General-purpose technologies—such as electricity, the internet, or now AI—can raise aggregate productivity while producing uneven outcomes across firms. Winners and losers often become clear only with a lag. Banks, however, must make forward-looking judgments continuously, because credit decisions cannot wait for the next multi-year wave of firm-level data.

Why banks’ interpretations matter

For private companies in Europe’s bank-centered financial system, the implications can be immediate. A lower PD typically signals lower expected credit losses and reduced regulatory or internal capital charges associated with a loan. In practical terms, that can translate into cheaper funding, larger facilities, or more willingness to refinance. Conversely, if PDs rise, lenders may tighten terms or ration credit—actions that can reshape investment decisions well beyond the firms directly affected.

The results also underscore that AI’s perceived benefits are not being allocated evenly. The reassessment in the study is weaker for labor-intensive firms—businesses where a larger share of costs and operations depend on human labor. That nuance suggests banks may view AI as less straightforwardly beneficial, or more socially and operationally disruptive, in settings where automation threatens established workflows, bargaining arrangements, or required reskilling. It may also reflect uncertainty about execution risk: deploying AI at scale in people-heavy processes can require costly organizational change, new controls, and careful compliance frameworks.

From narratives to numbers

A key contribution of the research is methodological: rather than inferring market beliefs from stock prices or venture investment, it uses banks’ own granular PD estimates. Those models blend hard financials with softer signals such as sector outlook, management quality, and macro assumptions. In other words, they are designed to convert narratives into risk parameters.

That conversion process is not purely mechanical. Banks have repeatedly been shown—in a wide body of finance research—to respond to policy shifts, crises, and new information by adjusting internal risk metrics, sometimes amplifying real-economy effects. The AI case adds a new layer: technology diffusion and attention cycles may now be influencing credit allocation through the risk models that sit at the heart of modern banking.

Still, a falling PD is not proof that AI will deliver durable profitability improvements for the affected borrowers. It indicates that lenders, collectively, are leaning toward a more optimistic scenario—at least relative to their pre-ChatGPT baseline beliefs.

What happens next

The immediate question for credit markets is whether these early PD revisions persist once AI adoption costs, governance needs, and legal constraints become clearer. Banks face pressure to balance competitive lending with model discipline, especially as supervisors scrutinize risk management and as firms experiment with AI in customer service, coding, marketing, and back-office functions.

For borrowers, the message is equally pointed: in AI-exposed industries, lenders may now reward credible readiness—data infrastructure, workforce plans, and risk controls—not just traditional balance-sheet strength. For labor-intensive firms, the weaker effect suggests they may need to work harder to convince banks that AI is an augmenting tool rather than a destabilizing shock.

Why it matters

Credit is a steering wheel for the real economy, especially in Europe where bank lending remains central for private firms. If banks systematically view AI exposure as risk-reducing, funding could tilt toward sectors perceived as AI “winners,” accelerating structural change. But if those beliefs prove premature—or overlook frictions in labor-intensive businesses—today’s PD cuts could become tomorrow’s credit mispricing.