Operational risk, cyber risk and resilience in the AI era
AI does not create a new risk taxonomy; it makes existing operational, cyber and third-party risks faster and more connected. Institutions stay resilient by mapping AI into critical services, enforcing limits in systems rather than prompts, monitoring key controls continuously, concentrating less on single AI providers, and testing fallbacks that work when AI or a vendor fails.
Which operational risks does AI amplify?
| Basel event type | How AI changes it | Priority controls |
|---|---|---|
| External fraud | Deepfake voice and video, AI-written phishing, synthetic identities | Call-back verification, phishing-resistant MFA, liveness checks |
| Execution, delivery and process management | Agents and automation execute errors at scale | System-enforced limits, approval gates, continuous control monitoring |
| Business disruption and system failures | Dependence on AI services and cloud providers | Fallback procedures, provider mapping, resilience testing |
| Clients, products and business practices | Chatbots give misleading advice; biased decisions | Guardrails, conversation sampling, fairness testing |
| Internal fraud | Model or prompt manipulation by insiders | Segregation of duties for model changes, logging, access reviews |
How should banks manage third-party and AI vendor risk?
Treat each material AI provider as a critical third party: due diligence on the model, contracts that ban training on your data and require change notice, monitoring of outputs, and an exit plan. Report how many material services depend on each provider. See the AI vendor risk guide.
What should an AI-era RCSA include?
Start from the function-level risk register, add each unit's AI use cases from the inventory, rate inherent and residual risk, and agree dated actions. The RCSA guide sets out the six steps and the rating scale; every register page offers a free workbook.
How do we keep control testing up to speed with automation?
Move the ten highest-value, data-rich controls to continuous control monitoring, route exceptions to owners with deadlines and report trends monthly. Keep periodic testing for the rest.
Register functions for this topic
Guides in this topic
AI cyber risk for banks: new threats and the controls that work
How AI changes cyber risk for banks: deepfake fraud, AI-written phishing, prompt injection and data leakage, and the controls that address each threat.
Third-party AI and vendor risk: due diligence, contracts and exit plans
How to manage AI vendor and third-party AI risk: due diligence questions, contract clauses, concentration risk and exit plans for banks and fintechs.
Sample operational risk controls for digital banking
Sample operational risk controls for digital banking: cyber, fraud, outages, third parties and privacy, from the Doorstep risk register, updated for AI.
How to start continuous control monitoring in a bank
How banks move from annual control testing to continuous control monitoring: choosing controls, data sources, thresholds, alerts and evidence.
How to run a risk and control self-assessment that covers AI
Step-by-step guide to a risk and control self-assessment (RCSA) that includes AI use cases, using a risk register to start and a workbook to record results.
Primary sources (reviewed 17 Oct 2026)
- Basel Committee: Principles for Operational Resilience (March 2021)
- Basel Committee: Revisions to the Principles for the Sound Management of Operational Risk (March 2021)
- Financial Stability Board: Sound Practices for Responsible Adoption of Artificial Intelligence, final report (15 October 2026)
- Regulation (EU) 2022/2554, Digital Operational Resilience Act (DORA)
- Federal Reserve, FDIC and OCC: SR 23-4, Interagency Guidance on Third-Party Relationships: Risk Management (June 2023)
This page explains supervisory guidance in plain language. It is not legal advice; confirm how each text applies to your institution and jurisdiction.
Frequently asked questions
How does AI change operational risk?
AI speeds up processes and errors alike. Automated and agentic processes can execute mistakes at machine speed, AI providers add concentration risk, and attackers use AI for deepfake fraud and phishing. The operational risk taxonomy stays the same; the controls need system-enforced limits, faster monitoring and tested fallbacks.
What is operational resilience in banking?
Operational resilience is the ability to keep delivering critical operations through disruption. The Basel Committee principles (2021) expect banks to identify critical operations, set tolerance for disruption, map dependencies including third parties, and test that they can stay within tolerance in severe but plausible scenarios.
Is AI a separate risk category?
Most supervisors treat AI risk as an extension of existing categories: model, operational, cyber, conduct and third-party risk. Manage it inside those frameworks, with an AI inventory linking each use case to the risks it affects.
What is the difference between operational risk and operational resilience?
Operational risk management aims to reduce the likelihood and impact of losses from failed processes, people, systems or external events. Operational resilience assumes disruption will happen and focuses on continuing critical services within tolerance.
Put these controls to work this month
Start on the cloud risk management software within 24 hours, or ask for training or advice tailored to your institution.