ESAs: financial institutions must act fast on frontier AI cyber risk
The European Supervisory Authorities (EBA, EIOPA and ESMA) urge financial institutions to adapt their cyber resilience to frontier AI models quickly and proactively. DORA and the AI Act provide the foundation; boards must set up governance without delay.
On 31 July 2026, the European Supervisory Authorities EBA, EIOPA and ESMA (the ESAs) published a joint statement on ICT risks from frontier AI models. Their message to financial institutions is clear: these models significantly accelerate cyber risk, and you need to act fast and proactively. For Dutch boards, this puts resilience against AI-driven attacks firmly on the agenda.
Why are frontier AI models a risk?
According to the ESAs, AI-enabled cyber tools could create systemic risk because they can:
- rapidly discover and exploit vulnerabilities;
- target vulnerabilities in shared infrastructure;
- leverage single points of failure across multiple entities.
The statement builds on the ESRB warning of 25 June 2026: if disruptions last several days or hit several institutions at once, the consequences could spread to the wider financial system and the real economy. The ECB has already asked the CEOs of significant banks to resolve open ICT findings from earlier inspections and the 2024 cyber resilience stress test without delay.
What role do DORA and the AI Act play?
The ESAs describe the existing EU framework, especially DORA and the AI Act, as a solid foundation.
- DORA remains highly relevant, the ESAs say, through the ICT risk management framework, testing, incident and recovery management, and ICT third-party risk management. This is the framework financial institutions themselves need to act on.
- The AI Act sets out a regime for ‘general-purpose AI models with systemic risk’. Providers of these models face additional obligations, including transparency, technical documentation and cybersecurity.
The ESAs stress that the framework is technology-neutral. But because vulnerability discovery and exploitation cycles are getting shorter, they urge institutions to act quickly.
Why does speed matter?
AI-driven attacks operate at machine speed. The ESAs note that periodic measures, such as an annual penetration test or audit, may therefore fall short. The statement and its annex set out three strategies:
- Prevention: up-to-date inventories of all IT assets (including APIs and AI components), security by design, a smaller attack surface, automated patching of high-risk systems, and enforcing cybersecurity standards across the supply chain.
- Detection: moving from periodic to continuous scanning and monitoring, with logging, behavioural monitoring and more frequent testing. SOC and red teaming capabilities can be enhanced with AI.
- Management: adapting incident response and business continuity to AI-assisted attacks and multi-system failures, protecting backups, and extending resilience testing to AI-enhanced scenarios.
Critical ICT third-party providers (CTPPs) are also in scope: the ESAs are embedding AI risks into their oversight and the 2027 examinations.
What does this mean for your board?
The ESAs place responsibility squarely with the management body. Financial institutions should establish governance for frontier AI risk “without delay”. Supervisors should ensure that boards are fully committed, with clear accountability, prepared response plans and sufficient investment. The Risk Appetite Framework should be reviewed, and board oversight should move from periodic to continuous, informed engagement.
Proportionality applies (DORA Article 4): measures should match your organisation’s size, risk profile and complexity.
Practical takeaway
Start with three questions. Have all open ICT findings been resolved? Has your board explicitly included frontier AI risk in the Risk Appetite Framework? And are your testing, patching and monitoring cycles fast enough for attacks at machine speed? Want to prepare your board? Take a look at our DORA boardroom training.
