On 7 July 2026, the European Central Bank directed Eurozone banks to have a plan in place by 31 October 2026 to address AI-enabled cyber threats capable of disrupting financial services. The mandate covers banks across the following countries in Europe: Austria, Belgium, Bulgaria, Croatia, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Portugal, Slovakia, Slovenia, and Spain.
It’s important to note that this isn’t just a guideline—it’s a deadline to take action against increasingly sophisticated cyber threats, and identity is one of the few control layers a bank can act on directly, without waiting on a vendor roadmap or an infrastructure overhaul.
Why the ECB is acting now #
This new mandate by the ECB underscores a larger trend toward stronger identity verification as organizations face increasingly sophisticated cyber threats accelerated by AI. In fact, identity has become the primary attack surface for cybercriminals, with 86% of phishing attacks now being AI-driven.
The mandate reflects a shift regulators can no longer treat as theoretical: AI has compressed the gap between when a vulnerability is discovered and when it’s exploited. Security teams used to have weeks to patch a flaw or catch a suspicious login. In AI-accelerated attacks, that window is now measured in just minutes.
There are three primary trends driving this compression:
Frontier models can find and exploit vulnerabilities independently. The latest models have demonstrated that they can find previously unknown vulnerabilities and build sophisticated attacks based on their findings. Additionally, they have shown the ability to combine previously moderate to low vulnerabilities into successful attacks.AI-generated phishing at scale. Attackers use generative AI to produce personalised, convincing phishing content faster and in far greater volume than manual campaigns ever allowed. An agent could use stolen credentials to gain access to systems with unpatched vulnerabilities.Credential theft to account takeover in minutes, not days. AI has the ability to build exploits based on vulnerabilities autonomously resulting in very short response times for security teams to respond. If a frontier model finds a previously unknown vulnerability, the challenge to respond is even greater. With stolen credentials, the AI now has the access it needs to perform the attack based on the newly found vulnerabilities.
Access management is a key control to combat the new reality we live in. As is already the case, legacy authentication mechanisms that rely on a person to make a judgment call under pressure—approving a push notification, reading back a code or trusting a voice on the phone—are no longer sufficient to protect access.
Where to start: the identity controls AI has made non-negotiable #
The ECB’s directive is outcome-based, not a prescriptive technology checklist. Banks must go beyond just a policy statement, and create a documented action plan that shows they’ve assessed their exposure to AI-enabled threats, and have concrete mitigations underway. In practice, that means addressing vulnerability management, automated threat detection, and identity and access controls with the same urgency the attackers now operate at.
Identity sits at the center of that plan for a simple reason: it protects the access to systems that could have an exploitable vulnerability. Additionally, it’s the control layer banks can harden fastest, and it’s the one AI is most directly built to attack.
Two areas carry the highest immediate and long-term impact, requiring minimal effort:
Phishing-resistant authentication for privileged users. Codes, links, and push approvals all rely on a human spotting a fake. Hardware-backed FIDO2/WebAuthn authentication removes that dependency—there’s no code to phish and no shared secret for AI to extract, because the credential is bound to a physical device rather than something a person can be tricked into handing over.*A company cannot stop at privileged users.*First off, the definition of privileged user can be difficult to define as many users have access to sensitive data and systems. Secondly, attacks commonly follow a pattern of gaining a foothold with a lower-privilege account and finding compromises to gain access to a higher-privileged account. To be fully protected, all users should be using phishing-resistant authentication solutions.Identify and protect public-facing access points and the help desk. Attackers will focus on the avenues they have access to, and that starts with access from the internet. Ensure all end points are known and require phishing-resistant authentication mechanisms. Another avenue that is easily accessible is the help desk, where AI tools can be used to convince a help desk agent to hand over access. If account recovery can be approved by a voice on the phone, it can be approved by a cloned voice. Help desk verification of users must include a step that an AI-generated voice or message can’t complete on its own.
The data backs the urgency: IBM’s threat intelligence research points to a 44% year-over-year increase in exploitation of public-facing applications, much of it via AI-assisted credential harvesting. Meanwhile, organizations that replace legacy multi-factor authentication (MFA) with FIDO2/WebAuthn authentication like hardware passkeys have seen account takeovers drop by as much as 99.9%—largely because there’s no phishable secret left in the flow to steal.
Continuous monitoring, just-in-time access segmentation, and incident response playbooks rehearsed against AI-accelerated timelines round out a complete plan. We’ve mapped all five identity controls banking security and risk teams should have in place—with the specific gaps AI opens in each—in a working checklist built for this deadline: Get the Checklist.
The DORA connection #
Banks already working through DORA’s ICT risk management and strong authentication requirements will recognize a lot of overlap here. Deploying phishing-resistant MFA to satisfy the ECB’s AI mandate simultaneously checks boxes for DORA’s authentication provisions. One identity investment addresses two regulatory obligations.
FAQ #
Which banks does the ECB’s AI cyber mandate apply to?
Eurozone banks operating in Austria, Belgium, Bulgaria, Croatia, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Portugal, Slovakia, Slovenia, and Spain.
What’s the deadline?
31 October 2026. The directive was issued 7 July 2026.
Is this the same as DORA?
No, but the two overlap significantly. DORA governs broader ICT risk management and resilience; this mandate specifically requires a plan for AI-enabled cyber threats. Phishing-resistant authentication satisfies requirements under both.
Where should a bank start?
Identity—specifically privileged-access authentication and helpdesk/call-centre verification—because it’s the fastest control to harden and the one AI most directly targets.
Next steps #
The full window between now and 31 October is short for an infrastructure overhaul, but identity controls don’t require one. Deploying hardware-backed, phishing-resistant authentication can happen inside existing IAM infrastructure, without a lengthy development cycle.
Start with the Identity Controls checklist to map where your bank stands today against the ECB’s expectations, and where to focus before the deadline.