Canada’s AI Cyber Direction: An Assume-Breach Checklist

The Government of Canada’s cyber security readiness direction, effective October 1, 2026, requires covered departments and agencies to submit a prioritized list of critical services in the Treasury Board of Canada Secretariat’s Enterprise Portfolio Management tool by November 30, 2026. Its required actions cover inventories, patching, identities, network containment, monitoring and recovery. Together, they give departments a concrete way to operate on an “assume breach” basis.

For a covered department, the implementation task is to connect each priority service to the systems, people and non-person identities that support it, then retain evidence that the relevant controls work. The Cyber Centre’s frontier-AI guidance recommends faster patching, less exposure, stronger authentication, continuous monitoring and zero-trust controls for organizations more broadly. Those recommendations help explain the threat; the departmental checklist below follows the federal direction.

Identify critical services and their exposure

Begin with the departmental critical-services inventory and rank services by the operational, legal, financial or reputational harm a compromise could cause. A service owner should identify the applications, infrastructure, devices, integrations and data on which each priority service depends. The direction allows phased implementation, beginning with the highest-priority services and extending the same controls to the rest.

The IT owner should keep each supporting application current in the Application Portfolio Management tool, including its owner and software version. An architecture diagram should show edge devices, internal segments, identity systems, management interfaces and administrative paths. Maintain a separate inventory of Internet connections, public-facing addresses where available, subnets and sub-domains, and provide it and later updates to the National Cyber Threat Notification System. For department-developed software supporting critical services, retain software bills of materials. Endpoint onboarding records from Shared Services Canada’s Endpoint Visibility and Awareness solution complete the initial view of managed and potentially rogue devices.

Accelerate patching and document delays

Make vulnerability work traceable from alert to decision. The security team should record relevant vendor, Cyber Centre and federal exposure notices; the affected assets; their Internet exposure and service criticality; the assigned owner; and the chosen response. The patch process must follow Government of Canada recommended time frames, with faster targets and shorter testing cycles for critical, Internet-facing and other high-risk systems. A single blanket deadline would obscure that risk-based requirement.

When timely patching is infeasible, record the compensating control and why the fix was delayed. Unsupported technology that cannot be adequately secured needs a replacement, isolation or retirement path. Keep remediation status and the decision on any remaining risk alongside the alert. The direction also calls for safe, responsible use of defensive AI tools to find vulnerabilities and review code; an inventory of covered repositories and resulting findings can show where that work is occurring. Its appendix offers notional progress indicators, rather than a new universal patch deadline.

Account for privileged users and AI agents

Identity controls must cover ordinary users, administrators and non-person entities. Enforce multifactor authentication for all user accounts and phishing-resistant methods for administrative, privileged and other high-risk accounts. Use centralized authentication and single sign-on by default, enforce least privilege, and review access regularly, starting with accounts that administer critical services. Dedicated administrative accounts should be confined to administrative tasks and protected by centrally managed privileged access management with auditable, just-in-time elevation. Where that system is unavailable, the direction permits an enterprise-grade password manager as an interim measure.

AI agents belong in the identity inventory alongside service accounts, service principals and machine identities. Record an accountable owner, business purpose, life-cycle state and privileges for each; review those privileges and minimize standing access and static credentials. Identity exports, approval records and access-review results provide evidence of that control. Centrally managed identity threat detection should also identify attack paths toward critical services. Staff simulations should cover AI-enabled phishing, voice phishing, deepfakes and helpdesk impersonation, with attention to people in high-risk roles.

Contain a foothold before it reaches a critical service

Reduce edge devices, remote access, administrative interfaces, Internet-facing services and APIs to essential functions. Place systems supporting critical services in segmented network zones. In their hosting environments, block outbound Internet traffic by default and allow only approved connections. A current zone diagram, approved outbound rules and configuration checks should let a reviewer see which routes remain open and why.

Vendor access to department-operated critical-service systems needs its own inventory of channels, owners and approved methods. Limit each session by time or task, log and monitor it, and remove dormant access promptly. Authorized administration must use government-approved hardened endpoints and protected, monitored channels isolated from lower-trust environments. Session records and access approvals can show whether those restrictions operated in practice, including where a provider manages part of a service on the department’s behalf.

Show that activity can be detected

Prioritize event logging for systems supporting critical services, with visibility across endpoints, identity systems, cloud components and edge devices. Protect logs from tampering with approved cryptographic safeguards and forward them to a central facility for storage, monitoring and analysis. Centrally managed endpoint detection and response must cover endpoints supporting critical services and administrative functions and feed departmental monitoring.

For each priority service, map the required log sources to the systems in its architecture diagram. Retain evidence that events reach the central facility and that endpoint protection is active and reporting. An assigned investigation owner and alert records make the monitoring chain visible from a compromised identity or device through to a containment decision.

Test containment and restoration

The departmental cyber security event management plan should identify decision authority, escalation paths and how the response connects to the business continuity plan. Run tabletop exercises against that arrangement. Approved incident playbooks should define key decisions, service levels, pre-approved containment actions and the telemetry to preserve for investigation. Exercise findings should lead to identifiable plan or playbook changes.

For systems supporting critical services, regularly test backup restoration and compare restored data with a known good reference. Keep those backups logically separate from production, use immutable or write-once storage where feasible, and protect the backup environment with access controls and monitoring. A useful readiness record pairs each restore result with the service it supports, the integrity check, the accountable owner and any unresolved gap. That evidence shows whether a department can restore the service after containment, rather than merely showing that backup copies exist.