Key Takeaways
- Regulators say technology is neutral, governance is not: supervisors care less about AI as a tool and more about how banks govern it and how it changes their risk profile.
- Nonbanks and private credit now hold around half of global financial assets, with growing interconnectedness to banks and markets.
- IMF and Harvard research show how nonbank systemic risk can transmit through bank credit lines, liquidity mismatches and opaque leverage.
- AI governance in banking should be designed with these systemic‑risk lessons in mind, not just with a narrow model‑risk lens.
- The strategic question is: are banks designing AI governance to withstand the same kind of systemic stress that nonbanks could trigger?
AI Governance in Banking
AI governance in banking is the set of policies, controls and decision‑rights that ensure AI systems are used safely, fairly and in line with prudential objectives. It covers model design, data, monitoring, explainability, accountability and escalation, and links AI usage directly to the bank’s risk profile and supervisory expectations.
Nonbank Systemic Risk
Nonbank systemic risk is the risk that stress in nonbank financial intermediaries—such as funds, insurers and private‑credit vehicles—spreads to banks and markets, amplifying shocks. It stems from leverage, liquidity mismatches and interconnectedness, including through bank credit lines and overlapping portfolios, in a sector that now holds around half of global financial assets.
What the ECB Means by “Technology Is Neutral, Governance Is Not”
In his February 2026 speech, the ECB’s Pedro Machado noted that AI is no longer on the “long‑term horizon” but at the core of banking, and that the supervisory approach is technology‑neutral. Supervisors do not regulate AI as such; they supervise how banks apply technologies, how they are governed and how they affect risk profiles, whether the innovation is AI, tokenisation or something else.
The ECB’s objective is not to slow innovation but to ensure banks embrace AI prudently, integrate it into business and digital strategies and remain “fully in control” before AI usage reaches systemic scale. With the right governance, AI can strengthen banks and support financial stability; with weak governance, it can amplify existing vulnerabilities.
What Nonbank Systemic Risk Is Teaching Supervisors
The IMF’s October 2025 work on nonbanks shows how quiet conditions can hide large, system‑wide vulnerabilities. Nonbanks now hold around half of the world’s financial assets, and in the US and euro area many banks have nonbank exposures exceeding their Tier 1 capital.
Stress scenarios where nonbanks become riskier and fully draw bank credit lines reveal how quickly nonbank stress can hit bank capital: about 10 percent of US banks and 30 percent of European banks (by assets) would see regulatory capital ratios fall by more than 100 basis points. This is systemic risk driven by system design—who is connected to whom and through what channels—rather than by one obviously weak institution.
Private Credit: A Nonbank Case Study in Governance Gaps
The Harvard Kennedy School paper “Private Credit & Systemic Risk” describes private credit as having “rapidly expanded” into an established source of loans to middle‑market firms and a key tool for private‑equity financing. It notes that private‑credit markets are less transparent, less liquid and more reliant on structured vehicles, making their risks harder to evaluate than those of traditional credit intermediaries.
The authors recommend that regulators and central banks expand the regulatory perimeter to include significant private‑credit funds, improve reporting for funds and borrowers, and integrate private‑credit dynamics into macroprudential policies such as countercyclical buffers. In other words, effective governance of private credit requires system‑level visibility and tools, not just firm‑level risk management.
AI Governance in Banking: Are We Repeating Nonbank Mistakes?
Industry commentary argues that 2026 is becoming a defining year for AI governance in banking, as regulators move from principles to more concrete expectations. The ECB and others stress explainability, continuous monitoring and clear human accountability for AI decisions that affect credit, fraud, pricing and compliance.
However, the nonbank experience suggests two additional questions for AI governance:
- Interconnectedness: How many banks rely on the same external AI and cloud providers, and what happens if a widely used model fails or is compromised?
- System design: How will AI‑driven decisions interact with nonbank shocks—for example, in how credit lines to funds are monitored, or how risk‑appetite changes in response to stress signals from models?
If governance frameworks focus only on model accuracy inside one institution, they may miss how AI behaviour could synchronise reactions across many banks or interact with nonbank stress.
A Simple Framework: Bringing Nonbank Lessons into AI Governance
- Map external dependencies
Identify critical AI and cloud providers and shared models used across the bank’s key risk and business functions; assess concentration and substitution options. - Trace links to nonbanks and private credit
Map where AI‑driven decisions influence exposures to nonbanks, such as credit‑line management, collateral haircuts or portfolio‑level de‑risking. - Integrate AI into macroprudential thinking
Consider how AI‑driven responses might behave in systemic stress scenarios—could similar models across banks trigger simultaneous tightening or fire sales? - Align board‑level oversight with systemic risk
Ensure AI governance at board and senior‑management levels explicitly covers systemic‑risk channels, not just reputational or compliance concerns. - Close data and reporting gaps
Just as Harvard and the IMF call for better data on private credit and nonbanks, banks should improve metrics and reporting on AI usage, performance and incidents to support supervisory and macroprudential oversight.
FAQ: AI Governance, Nonbanks and Systemic Risk
Why are regulators saying “technology is neutral, governance is not”?
Because they focus on how AI is governed and how it affects banks’ risk profiles, not on the technology itself; good governance can make AI stabilising, poor governance can make it destabilising.
How does nonbank systemic risk relate to AI governance?
Nonbanks show how system‑wide risks emerge from connections and behaviours; AI governance that ignores these system‑level dynamics may inadvertently reinforce the same vulnerabilities, especially where many banks use similar models or vendors.
What should banks prioritise in AI governance in 2026?
Banks should integrate AI governance into core risk frameworks, map external dependencies and nonbank links, ensure explainability and accountability, and be ready to demonstrate control and resilience under stress to supervisors.
Add Comment