AI cyber risk for banks: new threats and the controls that work
AI changes bank cyber risk in two directions. Attackers use it for convincing phishing, deepfake voice and video fraud and faster vulnerability discovery; and the bank's own AI adds attack paths such as prompt injection, data leakage and model manipulation. The response is to extend existing controls: call-back verification that never relies on voice or video alone, AI-specific testing, least privilege for AI tools, and monitoring of AI services.
The FSB's October 2026 sound practices list AI-related cyber and ICT risk as a distinct practice, to be managed inside the existing cyber framework rather than beside it.
Which AI-driven threats matter most for banks?
| Threat | Example | Key controls |
|---|---|---|
| Deepfake impersonation | Cloned voice of a CFO instructs an urgent transfer | Out-of-band call-back to a known number; dual approval for payments; staff drills |
| AI-written phishing | Flawless, personalised emails at scale | Phishing-resistant MFA; email authentication; awareness using real examples |
| Prompt injection | Malicious text in a document hijacks an internal assistant | Input and output filtering; least privilege; red-teaming |
| Data leakage via AI tools | Staff paste customer data into public tools | Approved tools; DLP on AI services; acceptable-use standard |
| Model and data poisoning | Training data manipulated to weaken a fraud model | Data lineage, integrity checks and drift monitoring |
| Faster exploitation | Vulnerabilities found and exploited quickly | Shorter patch cycles; asset inventory; continuous scanning |
Do we need a separate AI security policy?
Usually not. Add AI-specific requirements to the information security policy, the acceptable-use standard and secure development standards, and record AI services in the asset inventory. A separate document tends to drift away from the controls people actually follow.
Primary sources (reviewed 17 Oct 2026)
- Financial Stability Board: Sound Practices for Responsible Adoption of Artificial Intelligence, final report (15 October 2026)
- NIST AI 600-1: Generative Artificial Intelligence Profile (July 2024)
- Basel Committee: Principles for Operational Resilience (March 2021)
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
What is the best control against deepfake payment fraud?
A mandatory call-back to a pre-registered number and dual approval for payment instructions, so no instruction is executed on the strength of a voice or video call alone.
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.