Frontier AI and DORA: AI Service Providers to European Financial Entities Take Note

We last wrote about the EU Digital Operational Resilience Act (DORA) in our alert, “EU Commission Regulations on Digital Operational Resilience: A Reminder That DORA is Less Than Three Months Away and Will Apply to US and UK CTPPs,” shortly before it began to apply in January 2025. On 31 July 2026, the Joint Committee (JC) of the European Supervisory Authorities (ESAs)¹ issued a statement calling for a cross-sectoral, risk-based, and consistent supervisory approach to mitigate the Information and Communication Technology (ICT) risks — the risks which DORA seeks to address — stemming from frontier artificial intelligence (AI) models.²

The statement acknowledges and follows from the European Commission’s “EU Action Plan on Cybersecurity and Artificial Intelligence,” published 7 July 2026 (“Action Plan”). It seeks to translate the European Commission’s strategic objectives under the Action Plan into supervisory expectations for regulated EU financial entities, such as banks, broker-dealers, managers, and insurers.

The statement is not directed at and does not impose new legal obligations on the third-party providers of ICT services, including frontier AI models. It is important, however, because it shines a light on how regulators will expect EU financial entities, which are the customers of ICT service providers, to assess frontier AI models used or provided as part of their ICT services.

Moreover, the statement reinforces the contractual obligations that the financial entities are required to impose on ICT service providers under DORA. As we have previously noted, these include obligations connected with incident management and notification, business continuity and disaster recovery, data access, portability and return/deletion, termination rights, cooperation with regulators, audit and inspection rights, and subcontracting arrangements. If a financial entity determines that ICT services support a “critical or important function,” DORA imposes enhanced obligations, which include participation in threat-led penetration testing and further subcontractor obligations.

The effect of the statement is that developers and suppliers of frontier AI (more on this definition in the following section) may find that they are not only expected to demonstrate that their technology and services generally support their customers’ operational resilience obligations under DORA but also respond to the specific points raised in the statement, which we discuss in this alert’s “Points Raised in the Statement” section.

As we have seen during the last few years of acting for service providers facing the contractual demands of EU financial entities under DORA, service providers are well served by understanding the regulatory burdens and the limits of those burdens on those EU entities. This is important even if those service providers are unregulated technology companies and outside the EU.

Frontier AI?

A threshold question is whether the statement will apply to the AI ICT services provided.

The statement follows the Action Plan definition of “frontier AI models” as “the most advanced AI models available or under development” and describes frontier AI systems as those capable of performing a broad range of tasks that “approach, reach or exceed the current state of the art.”

In the Action Plan, the European Commission recognises that these models present a dual challenge. Frontier AI can significantly enhance cyber defence by improving threat detection, vulnerability management, and incident response. At the same time, increasingly capable AI models may also enable attackers to automate cyber reconnaissance, accelerate vulnerability discovery, and execute sophisticated attacks at unprecedented speed and scale.

Although these observations apply across sectors, they have particular significance for financial services, for which digital operational resilience is the core regulatory objective under DORA.

Points Raised in the Statement

In addition to DORA, the statement references the EU AI Act, which we covered in our recent alert, “Not Delayed, Not Deferred: EU AI Act Transparency Obligations Are Now in Force.” Based on the obligations arising from both DORA and the AI Act, the JC of the ESAs encourages financial entities, taking into account the expectations of their member state regulators, to adjust ICT risk management processes, procedures, and controls according to the following three risk mitigation strategies, with examples of measures set out in the statement’s Annex:

• Prevention that relies on comprehensive and continuously updated inventories of all information technology (IT) assets (including infrastructure, applications, data repositories, APIs, and AI/machine learning components); secure by design principles to ensure that systems are designed with built-in safeguards, rather than relying solely on reactive controls; strong monitoring and proactive patching processes designed to maintain resilience against evolving threats and reduce the window of exposure; and assessment of the risk from dependencies among IT assets
• Detection with systems and processes that scale up existing vulnerability discovery processes to match the existing threats; identify, analyse, and respond to intrusions before they escalate into significant incidents; are continuous, to reduce detection and response times, and enhance visibility of unusual or malicious behaviour; and make use of security operations centre management and red teaming capabilities with enhanced AI solutions added to existing technical and staff capabilities
• Management of cyber risk to improve operational resilience through operational resilience testing, enhanced disaster recovery, data backup capabilities, and increased cyber maturity; risk management frameworks, testing methodologies, and governance structures adapted to address AI-assisted threats and potential multisystem failures; management bodies working in parallel to ensure that accountability keeps pace with emerging risks and cybersecurity awareness evolves toward a more proactive and adaptive mindset; and the amendment of existing IT risk management frameworks

Translating These Strategies Into Financial Entity Expectations of Service Providers

In light of these risk mitigation strategies, frontier AI ICT service providers should expect financial entity customers to ask increasingly detailed questions about their cybersecurity architecture:

• governance of model development and deployment;
• resilience of supporting infrastructure;
• transparency regarding model limitations and failure modes;
• incident management and reporting processes;
• dependency on subcontractors and cloud infrastructure and pass-through of obligations; and
• business continuity and operational resilience arrangements.

As our experience already bares out, these requests are unlikely to arise not only from procurement or commercial contracts teams but also compliance governance and ICT risk management teams. The commercial due diligence applied to frontier AI providers is, therefore, likely to resemble the maturity that we have already seen in cloud outsourcing and other critical ICT services.

Proportionality in Approach?

As is the case with the application of DORA and similar financial services measures, financial entities should not be required to take a one-size-fits-all approach. As the JC of the ESAs notes in its statement, referring to Article 4 of DORA, in designing their risk mitigation strategies around the three pillars of prevention, detection, and management, financial entities will be required to take into account their size and overall risk profile, as well as the nature, interconnectedness, scale, and complexity of their services, activities, and operations.

Proportionality remains a key concept and, commercial realities permitting, service providers should ensure that financial entities observe this concept when seeking to impose contractual obligations on them in the name of regulation.