On July 7, 2026, the European Central Bank (ECB) sent a letter to the CEOs of all banks under its supervision, instructing them to develop a formal action plan against AI-based cyberattacks and submit it to their supervisory team by October 31.
The letter bears the handwritten signature of Supervisory Board Chair Claudia Buch and details the cyber risks of Frontier AI, as described in previous regulatory playbooks. The scope extends beyond banks with "Europe" in their name to include subsidiaries of JP Morgan Chase, Goldman Sachs, Citi, and Morgan Stanley in the Eurozone, which are also under ECB supervision and therefore subject to the same deadlines as banks in Frankfurt and Paris. According to a report by American Banker (a long-established independent specialist media outlet based in New York that focuses on the financial and banking industry), no US regulator has set similar requirements, while major US financial institutions are already receiving an overwhelming number of questions from regulators regarding similar risks.

Given these circumstances, it can be said that the ECB in Europe has the potential to become a leading example for the rest of the world, including Japan and the United States.

A noteworthy aspect of the letter is that the majority of the items the ECB is requesting relate to the software supply chain. This refers to third-party developed software, open-source components, and development pipelines—things that banks have incorporated into everything over the past decade.

Why now? The reason is attack code that can be built in just a few minutes.

The ECB frankly observes that the newly emerging AI models "can discover vulnerabilities and generate attack code at an unprecedented speed, shortening the time from vulnerability discovery to attack code generation." Buch views this as "not a temporary phenomenon or a risk tied to a specific tool, but a long-term shift in the threatening environment."
On the same day, the European Systemic Risk Board officially issued a warning that the Frontier Model can execute fully automated attacks and build attack code that works in "minutes or hours" where it would take a human "days or weeks," viewing this change as a "collapse of the time window for response." The Board's greatest concern is the scale of the damage. They believe that if a wave of threats exploiting vulnerabilities hits many financial institutions simultaneously, "there will be zero effective risk mitigation frameworks available."

And that hypothesis is gaining traction. Earlier this year, Anthropic reported that a tightly restricted frontier model had uncovered numerous previously unnoticed vulnerabilities in major operating systems and browsers. The UK's NCSC, citing an annex to the ECB letter, has urged security personnel to prepare for a "wave of vulnerability patches."
When the development of attack code is reduced from weeks to minutes, any process that relies on human speed (such as monthly patch cycles, quarterly dependency reviews, and yearly penetration testing) becomes vulnerable to failure at any moment.

What are the items that the ECB is actually requesting?

The action plan, due October 31, must include specific actions, the names of those responsible, allocated budgets and personnel, and implementation schedules. All of these are positioned to fulfill existing obligations under the Digital Operational Resilience Act (DORA), which we have previously addressed. The accompanying document outlines six key areas, three of which are directly related to application security and the software supply chain.

The first step in protecting your attack surface is knowing what assets you own.　

The letter identifies the identification of ICT assets, such as "software developed by third parties and open-source components," as a top short-term priority, explaining that this is "key to prioritizing corrective actions." Without a clear inventory, it's impossible to implement countermeasures against attacks, patching, or decommissioning. And for many banks, that's the open-source dependency tree beneath their applications.

Vulnerability management and patch management must be accelerated "on a large scale".

The ECB anticipates that if vendors, the open-source community, and even internally developed software can quickly identify and fix vulnerabilities, the frequency and volume of patching will improve. The ECB strongly recommends creating an environment for AI-powered security tools, prioritizing vulnerability scanning and including safety measures, human oversight, and robust risk management.

Security by design: Solving and reducing problems at the upstream stage.

In structural security measures, "practicing software development based on security by design leads to addressing vulnerabilities before deployment, not only in terms of meeting deadlines but also through proper consideration of secure code." The fewer defects there are after deployment, the fewer emergency patches will be needed. This is a countermeasure that works sustainably against attack code that attacks around the clock.
The same applies to other areas (monitoring and detection, governance and supply chain assurance, and operational resilience). The Board must verify whether the ICT budget, personnel, and tool infrastructure will be beneficial for accelerating patching, and financial institutions must understand the "accelerating vulnerability disclosure and patching readiness" of ICT providers.
The ECB has changed the deadline for submitting its annual IT Risk Questionnaire from September 2026 to February 2027 to allow banks to focus on this initiative.

Gap between requirements and response time

For a security team, the question is, just how large a scale does "significantly accelerate patch management" actually mean?

Our company recently analyzed over 1,000 AI agent attack sessions from actual intrusion incidents. The results were consistent with the ECB's assessment. In other words, instead of attackers exploiting unique zero-day vulnerabilities discovered by AI, they weaponize infrastructure that exposes known and publicly available CVEs at extremely high speeds, 24/7 without interruption. Furthermore, even if they have the capability to block entry points and risks when they occur, they lack the time to do so.

This is the issue of "triage," or prioritization, that the ECB is now beginning to mandate at the Governing Council level. A typical enterprise application is largely comprised of code written outside the organization (open-source packages, transitive dependencies, vendor-provided components), with a backlog of thousands of items. As the time between vulnerability disclosure and exploitation shortens, "addressing it in the next sprint" becomes a decision to accept that risk. The ECB is calling for the Governing Council to formally assume responsibility for making that decision and to incorporate metrics and tolerance limits into the risk tolerance framework.
As the ECB states in its letter, this is not something that can be handled by humans alone. That is why it recommends prioritizing AI-powered measures and scanning, rather than simply urging banks to make an effort.

Responding to requests: Automate what can be automated.

Here, we'll show how modern application security platforms are directly linked to the ECB checklist items.

Understanding your company's attack surface, including open source.

Mend.io builds a complete and continuously updated inventory of all open-source and third-party developed components within an application, including transitive dependencies, and represents them as a standards-compliant Software Bill of Materials (SBOM). This fulfills the requirement initially outlined in the letter (identification of assets as the basis for prioritized corrective actions) and provides supply chain assurance teams with the evidence required by DORA reviewers.

Implement the prioritization required by regulatory authorities.

The ECB is calling for prioritizing vulnerability scanning because the total number of CVEs has reached an unmanageable level. Mend.io's reachability analysis distinguishes between vulnerabilities hidden in the code an application runs and vulnerabilities in paths that haven't been used for a long time. This allows teams to use the reduced time spent on remediation to discover AI-powered entry points that attackers can actually reach.

Performing patch updates at extremely high speed

"Handling large volumes of patches" is precisely why dependency automation exists. Mend Renovate automatically detects outdated and vulnerable dependencies, creates tested new pull requests, and keeps large repositories up to date without engineers having to spend time on it.
This transforms the wave of patches from a crisis of understaffing into a process that runs automatically in the background. This is exactly what the ECB describes as an automated proposal—"AI-based tools…with appropriate measures [and] human oversight"—which is then recorded after human approval.

Check security before deployment.

The security-by-design approach described in the letter requires implementation in the pipeline, not in the production environment. Mend SAST and container scanning detect vulnerabilities in your own code and images before release. Mend AI extends the same discipline to the modern attack surface: AI models, agents, and AI-generated code embedded in applications.

The above items do not cover all of the ECB's requirements. Decommissioning outdated core systems, network segmentation, and crisis management training are infrastructure tasks that the AppSec platform cannot handle. However, the software supply chain aspect of the requirements can be automated. Automating this would free up personnel resources for areas that cannot be automated.

Things to do before receiving a letter from regulatory authorities

The OCC has already stated that AI is "significantly changing the cyber threat landscape." The Bank of England, the FCA, and the UK Treasury issued a joint statement in May on frontier AI and cyber resilience. FS-ISAC has published a warning to the industry about the discovery of vulnerabilities using AI. The countries participating in Five Eyes have also issued a joint statement. The letter from the ECB not only provides information on submission deadlines, but Mr. Buch has already promised to release further updates on post-quantum cryptography.
Financial institutions under ECB supervision have a deadline of October 31st for submitting their action plans. This means that you need to start building and prioritizing your inventory and laying the groundwork for automation now. Even financial institutions not under ECB supervision should take this as a heads-up.
In any case, as the ECB listed first, the first step is to have a precise understanding of your own software and its components.