In response to ECB mandate SSM-2026-0301
The Identity Side of the
ECB AI Cyber Risk Mandate: A Guide for CISOs and
IAM Leaders
How to close the identity gaps the ECB letter names before October 31, 2026
This guide is for CISOs and IAM Leaders at ECB-supervised significant institutions responding to the ECB AI cyber mandate (ECB letter SSM-2026-0301) who are now asking: what do I actually need to do on the identity side, and in what order?
We've pulled the identity requirements from the ECB AI cyber mandate into one place, mapped each one to specific controls you can deploy now, and separated what the mandate requires from what we recommend—and it's all structured around the attack patterns frontier AI models actually use in production environments.
We recommend reading this guide alongside our main briefing, The ECB Just Made AI Cyber Risk a Board-Level Emergency: Here’s Your Playbook.That document covers the full ECB AI cyber mandate, all 6 ECB focus areas, and 7 recommended operational tracks. This guide zooms in on the identity actions your IAM team needs to execute across those tracks.
Part 1: The Mandate
Identity Security controls against AI attacks: What actually works
Silverfort’s research team tested frontier AI models, including Anthropic’s Mythos, against real enterprise environments. The most important finding wasn’t a new vulnerability or exploit technique. It was time. AI compressed the entire attack chain—from reconnaissance to lateral movement and privilege escalation—into minutes. At the same time, it became clear that identity had become the primary battlefield and the primary control plane. Rather than relying on novel exploits, these models consistently abused trusted identities, over-privileged accounts, legacy authentication protocols, and Active Directory trust relationships to move through the environment at machine speed.
This fundamentally changes which security controls remain effective.
Admin-time controls
Controls designed to operate before an attack, like vulnerability management, posture management, and periodic identity governance, remain essential but can no longer eliminate risk fast enough.
Post-access controls
Controls that operate after access is granted (like detection and response) are often too late once an AI-powered attacker begins moving at machine speed.
Runtime controls
What consistently proved effective was controls that act at runtime: inline identity enforcement at the moment of authentication, before access is granted.
This is also why the ECB’s expectations are significant. Continuous verification, Least Privilege, MFA, identity segmentation, and compensating controls for legacy systems are no longer simply identity best practices. They are the controls most likely to stop AI-powered attacks from spreading once an attacker gains an initial foothold.
Practical insights
See how to stop AI attacks
Get the readiness guide for identity and security teams
How this guide maps to the ECB AI cyber mandate
The ECB's action plan spans multiple domains, from attack surface management to operational resilience. This guide translates those expectations into identity-focused operational workstreams, mapping each recommendation to the relevant ECB areas and providing practical actions security leaders can implement immediately.
| Guide Section | ECB Area | Checklist Items |
|---|---|---|
| Stop credential abuse (including privileged ones) | Area 1 (Attack surface), Area 2 (Patch management), Area 5 (Defense-in-depth & zero trust) | |
| Break lateral movement | Area 5 (Defense-in-depth & zero trust) | |
| Protect service accounts | Area 5 (Defense-in-depth & zero trust) | |
| Secure privileged access | Area 4 (Governance & supply chain), Area 5 (Defense-in-depth & zero trust) | |
| Restrict weak protocols | Area 5 (Defense-in-depth & zero trust) | |
| Monitor and detect identity threats | Area 3 (Monitoring & detection) | |
| Contain incidents | Area 6 (Operational resilience) |
How to use this guide
Each section distinguishes between:
What the ECB requires
Controls and capabilities explicitly required or implied by the ECB's supervisory expectations and the Digital Operational Resilience Act (DORA).
Our recommended approach
Additional measures based on Silverfort's field experience defending against frontier AI attacks. While not explicitly mandated by the ECB, these controls can improve resilience and strengthen banks' overall response plans.
01
Stop credential abuse at the source
ECB Areas 1, 2 & 5: Attack Surface, Patch Management & Defense-in-Depth
Why it matters
Frontier AI models start with the fastest route to privileged access. Techniques such as replicating the entire directory (DCSync), cracking service account passwords offline (Kerberoasting), and exploiting accounts without pre-authentication (AS-REP Roasting) allow attackers to turn weak identity controls into enterprise-wide compromise within minutes.
See the FAQ below for what each of these techniques means.
ECB-recommended actions
- Protect authentication paths to high-value and internet-facing assets.
- Apply compensating controls where systems cannot be patched immediately.
- Enforce Least Privilege and MFA as foundational controls.
Additional actions
- Restrict directory replication to domain controllers and approved sync accounts.
- Identify and remediate accounts that do not require Kerberos pre-authentication.
- Virtually fence service accounts to approved sources, destinations, and protocols.
- Extend privileged access protection beyond traditional PAM vault coverage.
How Silverfort helps
Silverfort enforces Identity Security inline with the authentication flow, allowing organizations to:
- Block unauthorized directory replication.
- Protect privileged accounts with Adaptive MFA/deny policies across all protocols.
- Apply Just-in-Time access instead of standing privileges.
- Virtually fence service accounts.
- Discover shadow admins and newly privileged accounts.
02
Break lateral movement
ECB Area 5: Defense-in-Depth & Zero Trust
Why it matters
AI-powered attacks chain legitimate authentications at machine speed. Every authentication looks legitimate in isolation. The danger comes from the sequence.
Static allowlists are no longer enough. The same admin logging in from a Privileged Access Workstation during business hours may be expected. The same credentials authenticating from an unknown workstation at 3am toward a domain controller should trigger an entirely different response.
This is exactly how attacks such as pass the hash (moving from a compromised workstation to a domain controller in minutes) or PsExec fan-out (using legitimate credentials to spread simultaneously across dozens of servers) succeed. Every authentication uses valid credentials and legitimate protocols, but together they form a malicious attack path that must be stopped.
ECB-recommended actions
- Continuously verify identities before granting access.
- Enforce MFA for privileged access.
- Apply identity-based segmentation and Least Privilege.
Additional actions
- Evaluate every authentication using real-time identity and risk context.
- Restrict privileged identities to approved authentication paths.
- Apply MFA on all admin protocols (RDP, WinRM, PsExec, SMB).
- Phase out or restrict weak authentication protocols to sensitive assets.
- Apply adaptive authentication policies that dynamically allow, challenge, or deny access based on risk.
How Silverfort helps
Runtime Access Protection (RAP) evaluates every authentication at runtime and applies adaptive policies inline to allow, step up, or deny access based on risk in all protocols—preventing AI-powered attacks from moving laterally before access is granted. This is what actually stopped Mythos.
03
Protect service accounts
ECB Area 5: Defense-in-Depth & Zero Trust
Why it matters
Service accounts are among the most attractive targets for AI-powered attacks. They often hold excessive privileges, cannot perform MFA, use long-lived credentials, and are rarely monitored. In many enterprise environments, they generate a significant share of authentication traffic, making them an ideal mechanism for lateral movement and privilege escalation.
ECB-recommended actions
- Continuously discover and inventory all service accounts and other non-human identities.
- Implement compensating controls where MFA cannot be applied.
Additional actions
- Baseline expected authentication behavior.
- Restrict every service account to approved sources, destinations, and protocols.
- Disable or tightly restrict dormant service accounts (90+ days no logon).
How Silverfort helps
Silverfort continuously discovers and classifies service accounts across hybrid environments. It baselines normal behavior and applies Virtual Fencing, restricting where and how service accounts can authenticate without changing credentials or disrupting business processes. Dormant accounts can be automatically identified for quick remediation.
04
Secure privileged access at scale
ECB Areas 4 & 5: Governance, Defense-in-Depth & Zero Trust
Why it matters
Privileged identities remain the fastest path to enterprise compromise. AI-powered attackers actively exploit credential coercion techniques such as PetitPotam, abuse cross-tier administration, and target privileged accounts protected only by passwords. Traditional PAM solutions often protect only a subset of privileged identities, leaving significant gaps across hybrid environments.
ECB-recommended actions
- Enforce MFA for all privileged accounts.
- Apply Least Privilege and continuous verification.
- Incorporate identity risk into governance and board reporting based on risk appetite framework
Additional actions
- Restrict where domain controller machine accounts can authenticate.
- Prevent cross-tier administration.
- Extend privileged access protection beyond traditional vault coverage.
- Protect privileged accounts with MFA across all protocols, including legacy.
- JIT access to remove standing privileges.
How Silverfort helps
Silverfort protects privileged access at scale and beyond the vault by discovering all privileged identities and enforces security controls inline with the authentication flow.
Runtime policies apply Adaptive MFA to Active Directory, legacy systems, and administrative protocols, isolate administrative tiers, restrict machine account authentication, and remove standing privileges with JIT access. It covers all privileged identities that Frontier AI targets first, including those traditional vaults cannot cover: Tier 0 (Domain Admins, PKI, Entra Connect), Tier 1 (Server Admins, DBAs, service accounts), Tier 2 (Helpdesk, IT Support).
05
Restrict weak authentication protocols
ECB Area 5: Defense-in-Depth & Zero Trust
Why it matters
ECB-recommended actions
- Modernize legacy technologies and authentication protocols where possible.
- Apply compensating controls where modernization is not immediately feasible.
Additional actions
- Inventory and quantify weak protocol usage.
- Prioritize restricting weak protocols to sensitive systems.
- Phase out NTLM and other legacy protocols over time.
- Track legacy protocol reduction as a security KPI.
How Silverfort helps
Silverfort provides complete visibility into legacy authentication protocols and enforces protocol-aware access policies inline at authentication. Organizations can selectively block or step up risky protocols without disrupting critical legacy applications, enabling phased modernization while immediately reducing exposure.
06
Detect identity threats in real time
ECB Area 3: Monitoring & Detection
Why it matters
ECB-recommended actions
- Strengthen authentication monitoring.
- Improve AI-enabled defensive capabilities.
Additional actions
- Baseline behavior across all identities.
- Detect credential abuse and authentication anomalies in real time.
- Feed identity signals into existing SIEM and SOAR platforms.
- Deploy honeytokens and tripwire identities to expose automated reconnaissance.
How Silverfort helps
Silverfort Identity Threat Detection (ITDR) continuously baselines authentication behavior and detects credential attacks, MFA fatigue, brute-force attempts, and anomalous identity activity in real time. High-confidence identity signals are integrated directly into existing SIEM and SOAR workflows, improving detection without requiring additional infrastructure. Risk-based response is generated automatically and inline to block risky access before it completes.
07
Contain incidents before they spread
ECB Area 6: Operational Resilience
Why it matters
ECB-recommended actions
- Integrate identity into crisis management and operational resilience planning and test them (aligned with DORA)
- Regularly test incident response and recovery procedures. ontrols where modernization is not immediately feasible.
Additional actions
- Predefine emergency identity containment policies ("big red button" scenarios): deny all auth from a compromised account, block weak protocols to Tier-0, enforce MFA on all privileged access.
- Automatically isolate compromised identities and privileged access paths.
- Integrate identity controls with EDR and SIEM signals.
- Validate that identity containment can be executed within minutes through regular exercises.
How Silverfort helps
Silverfort enables organizations to contain identity-based attacks in seconds by enforcing runtime policies inline with the authentication flow, including predefined emergency policies. Integrations with existing EDR/SIEM platforms enhance the attack signal and containment. Security teams can immediately block or challenge compromised identities, segment affected access paths, and activate predefined containment policies to stop attackers before they can continue moving through the environment.
12-Week Identity Security roadmap for ECB mandate compliance
The ECB's expectations span multiple teams and technologies, but they don't need to become multiple disconnected projects. Based on our experience helping organizations defend against AI-powered attacks, we've found that the highest-impact controls can be implemented in a structured sequence, delivering meaningful risk reduction within weeks while building toward long-term resilience.
The roadmap below prioritizes controls by risk reduction, implementation effort, and dependency, allowing banks to demonstrate measurable progress well before the October 31 submission.
Work with this plan to get 75% of identity requirements enforced by Week 8, and ready for JST submission at 93% by Week 12.
| Phase | Focus | Prioritize | Outcome |
|---|---|---|---|
| Phase 1 (Weeks 1–2) Quick wins | These controls can typically be implemented rapidly and immediately reduce the attack paths most frequently exploited by AI-powered attackers. |
| Immediate reduction of privileged credential abuse while establishing governance visibility. |
| Phase 2 (Weeks 2–4) Build the foundation | Before enforcing controls, establish visibility into your identity landscape. |
| A complete inventory of the identities and authentication paths AI-powered attackers are most likely to abuse. |
| Phase 3 (Weeks 4–8) Deploy core runtime controls | Once visibility exists, begin enforcing runtime controls at the authentication layer. |
| Identity becomes an active control plane capable of stopping AI-powered attacks before lateral movement begins. |
| Phase 4 (Weeks 8–12) Harden the environment | With runtime protection in place, reduce structural exposure across the environment. |
| A hardened identity infrastructure aligned with the ECB's defense-in-depth expectations. |
| Ongoing Reduce legacy exposure | Legacy protocols remain one of the most common attack paths exploited by frontier AI. |
| — |
By the end of the 12-week program, organizations should aim to have:
- MFA protecting all privileged identities
- Every access request is evaluated at runtime
- Runtime controls enforcing access decisions inline
- Service accounts discovered and fenced
- Identity threat detection integrated with the SOC
- Tier-0 isolated from lower-trust environments
- Emergency identity containment procedures tested
- Legacy authentication actively being reduced
Identity Security Roadmap: Checking the ECB box
Phase 1 | Weeks 1–2
Quick wins
- MFA on all privileged accounts
- MFA on admin protocols (RDP, WinRM, RsExec, SMB)
Highest-impact controls. Deploy in days, not weeks.
3 of 14 identity requirements
Phase 2 | Weeks 2–4
Foundation
- Discover & classify all service accounts
- Quantify & restrict NTLM to Tier 0
You cannot fence what you don't know.
6 of 14 identity requirements
Phase 3 | Weeks 4–8
Core controls
- Fence service accounts to expected envelope
- PAW enforcement for admins
- Deploy identity threat detection
Runtime controls enforced. SOC has identity signals.
10 of 14 identity requirements
Phase 3 | Weeks 4–8
Hardening
- DCSync restriction to DCs only
- Tier 0 isolation / cross-tier deny
- Dormant account restriction
Defense-in-depth hardened. Ready for JST submission.
13 of 14 identity requirements
Ongoing
Reduce legacy exposure
- Phase out NTLMv1
- Replace weak protocols
- Track reduction as KPI
Phase NTLM restriction broader.
14 of 14 identity requirements