The Canadian Centre for Cyber Security (Cyber Centre) is warning Canadian organizations that cyber threat actors are increasingly using automated and AI-enabled tools to identify and attack internet-facing systems. AI-enabled offensive security tools can automate or accelerate parts of an intrusion that previously required more direct human involvement. The Cyber Centre is tracking a growing number of these tools, including ARTEX, CyberStrikeAI and PentAGI.
This cyber security advisory describes how these tools speed up an intrusion. For the measures every organization should apply, see our cyber threat bulletin on agentic artificial intelligenceFootnote 1.
The threat to Canadian organizations
AI tools capable of supporting cyber threat activity are rapidly improving and becoming increasingly available. In our National Cyber Threat Assessment 2025-2026, we assessed that AI technologies are almost certainly lowering the barriers to entry and enhancing the quality, scale and precision of malicious cyber threat activityFootnote 2. Threat actors with limited technical expertise can use publicly available AI models for malicious purposesFootnote 3. This development builds on the widespread availability of offensive security tools (OSTs), which can provide threat actors with capabilities that would otherwise require greater technical expertise. More recently, we have observed OSTs incorporating AI to further support or automate aspects of cyber security testing and offensive activity.
Of the globally observed AI-enabled OST infrastructure identified through Cyber Centre analysis, approximately 46% was associated with ARTEX, 28% with CyberStrikeAI and 14% with PentAGI, with several other tools making up the remainder. While these tools vary in capability and maturity, their continued development demonstrates how AI could further reduce the technical expertise required to conduct malicious cyber activity.
On September 29, 2026, we issued a statement on reports of suspicious activity targeting publicly accessible websites, including Government of Canada websites. Those reports included suspected AI agent activityFootnote 4. Earlier that month, we warned that threat actors were using AI to find and attack internet-exposed industrial control systems. That campaign affected critical manufacturing, energy, water and wastewater, chemical, food and agriculture, and commercial facilitiesFootnote 5.
Public reporting in October 2026 described ARTEX as an autonomous AI-enabled penetration testing framework used against organizations abroadFootnote 6Footnote 7Footnote 8Footnote 9. The developer made the project closed source on October 8, 2026. By then it had been copied more than 12,000 times. Once a tool like this is released, organizations should expect it to stay in circulation.
The Cyber Centre is also aware of cyber threat activity targeting Canadian organizations from servers running ARTEX. This activity has been ongoing for several months and continues to target organizations across multiple sectors. The Cyber Centre has not attributed this activity to a specific threat actor. Targeting of this kind is not an indication of compromise but demonstrates that Canadian systems are within reach of these tools. When the Cyber Centre becomes aware of cyber threat activity targeting a Canadian organization, we notify the affected organization.
Separately, public reporting has linked ARTEX to state-aligned cyber threat activity. As noted in our bulletin on agentic artificial intelligence, Microsoft reported that a PRC actor tracked as Storm-0940 used the ARTEX framework against hundreds of organizationsFootnote 1Footnote 6. Other publicly reported ARTEX activity remains unattributedFootnote 8.
We assess that PRC cyber actors almost certainly pose the greatest ongoing cyberespionage threat to CanadaFootnote 10. The Cyber Centre and its partners have previously warned about the scale and persistence of PRC-linked cyber activity, including the use of compromised infrastructure, automated scanning and exploitation, and hands-on operator activityFootnote 11Footnote 12Footnote 13. These activities are distinct from the AI-enabled OST activity described in this advisory but demonstrate the broader threat environment in which increasingly capable automated and AI-enabled tools are becoming available.
Canadian organizations that run internet-facing applications, databases or cloud services should assume this activity could affect them. These tools can enable threat actors to identify and target systems based on what is exposed and vulnerable, increasing the potential exposure of organizations of all sizes.
How AI-enabled tools change an intrusion
Cyber-attacks are not yet fully autonomous. Cyber threat actors still choose the targets, set the objectives and oversee the operation; however, more of the work between those decisions is now done by machines. That gives defenders less time to detect and stop an attackFootnote 1.
Agentic AI has the potential to significantly reduce the time required to conduct an intrusion compared with human-led operations. As we noted in our Ransomware playbook, agentic AI can find weak points in a network, bypass defences, deploy malware and erase evidence of the intrusion, all in a single AI-managed activity stream. Reconnaissance and exploitation that once took days or weeks can now take minutesFootnote 14. Our guidance on social engineering describes the same emerging threat. Autonomous AI agents carry out entire attacks with little or no human involvement. They monitor results in real time and change approach when an attempt failsFootnote 15.
In some of the activity described in our bulletin on agentic artificial intelligence, less than three hours passed between selecting a target and gaining privileged access to cloud systems, including infrastructure, platform and software as a service (IaaS, PaaS and SaaS)Footnote 1. Organizations should plan for continuous scanning of internet-facing systems and assume that vulnerable systems could be identified, exploited and used to access data within hours rather than days. As we note in Frontier artificial intelligence, organizations should assume that AI-driven exploitation may bypass individual preventative controls in the absence of a defence-in-depth approach and significantly outpace vendors' capacity to publish corrective measuresFootnote 3.
These tools work by trial and error at high volume, rapidly adjusting after failed attempts. This activity can generate recognizable patterns across web access and error logs, application and application programming interface (API) logs, web application firewall (WAF) events, authentication logs, and firewall, proxy and intrusion detection system/intrusion detection and prevention (IDS/IPS) telemetry. As we noted in our bulletin on agentic artificial intelligence, much of this activity is currently widespread and noisy enough to be detected by sophisticated cyber defenders. We assess that AI-enabled cyber activity will likely become more targeted, less visible and harder to detect as the technology improvesFootnote 1.
What organizations should do
Our cyber threat bulletin on agentic artificial intelligence sets out the measures every organization should applyFootnote 1. They include phishing-resistant multi-factor authentication, limiting the access given to users, applications and service accounts, and reviewing access granted to users, which can include third party applications. They also include immediate patching of internet-facing systems, retaining security logs and monitoring for unusual sign-ins, new credentials and unexpected changes in access, deciding in advance who can isolate an account, application, network or service, practising response to incidents that unfold within hours rather than days, and separating important systems with tested offline backups. That bulletin also describes how critical infrastructure owners and operators can prepare to continue essential services during a serious cyber incident as per cloud service provided capabilities for monitoring. The measures below address the specific tradecraft in this advisory.
Monitor the cloud identity and application layer as per cloud service provided capabilities. Much of this activity takes place inside cloud identity and cloud application services, where it leaves little or no trace in an organization's own network trafficFootnote 6Footnote 7. Review the access granted to third-party and multi-tenant applications in your tenant and remove what is no longer neededFootnote 1. Enable alerts on a new secret or certificate added to an existing application, on newly granted application permissions, and on consent to an unfamiliar application. These survive a password resetFootnote 6. In authentication and audit logs, look for high rates of failed sign-in and account lockout across many accounts, repeated attempts from a changing set of source addresses, and a successful sign-in followed immediately by automated reads of mail, files or directory dataFootnote 16Footnote 17. Make sure refresh tokens are revokable, time-limited and scoped to the lowest duration acceptable to your user base, and that each application registration has its own credentials.
Harden the surfaces that turn a foothold into code execution as follows. Validate and parameterize all user input to prevent injection. Review file upload handling, including rename, move and extension logic, which can turn a benign upload into executable content. Mount application root directories as read-only wherever the application permits and disables runtime script compilation. On database servers, disable operating system command execution and automation features you do not need. Prevent changes to database logging configuration while the server is runningFootnote 18.
Suppress verbose error output in production. Detailed error messages and stack traces are what let an automated agent correct a failed payload and try again within secondsFootnote 18. Suppressing them directly reduces the tool's effectiveness.
Monitor for anomalous outbound activity. Monitor server and application tiers for unusual outbound connections, including connections to unexpected destinations, services or portsFootnote 19Footnote 20. Automated tools may connect to third-party services during exploitation or data exfiltration. Investigate outbound activity that is inconsistent with a system's expected function.
Look for the shape of automated activity, not just individual indicators. Signs include unusually high request volumes against a single application, long runs of failed requests followed by a success, repeated application errors followed immediately by a working request, and rapid bursts of read-only or validation-style API calls across many resource typesFootnote 16Footnote 17.
Verify your remediation. After any incident involving an internet-facing application or a on prem or cloud identity, re-scan your external surface and rotate every credential the actor could have reached. Then hunt for persistence that survives a password change. In cloud identity, that means credentials and permissions added to applications and altered account recovery addresses. Elsewhere it means new or modified service accounts, residual web shells, newly generated access keys and new scheduled tasksFootnote 21. Treat containment you have not verified as incomplete.
If your organization is adopting AI agents of its own, apply the same discipline to them. Our joint guidance on the careful adoption of agentic AI services explains how to scope agent permissions, monitor agent activity and contain an agent that misbehavesFootnote 22.
We encourage Canadian organizations to report cyber incidents and suspicious activity to the Cyber Centre at contact@cyber.gc.ca, and to consult the guidance linked belowFootnote 23Footnote 24.
Useful resources
Refer to the following resources for more information and useful advice and guidance.
Cyber Centre assessments and statements
- National Cyber Threat Assessment 2025-2026
- Cyber threat bulletin: Agentic artificial intelligence is accelerating cyber attacks (2026-10-25)
- Cyber threat actors use artificial intelligence in an active global campaign to disrupt internet-exposed programmable logic controllers (2026-09-08)
- Statement regarding reported activity targeting Government of Canada websites (2026-09-29)
- Cyber threat bulletin: People's Republic of China-sponsored cyber activity against Canadian provincial, territorial, Indigenous, and municipal governments
Joint and partner advisories
- Joint guidance on PRC government-linked cyber threat actors combining automated and hands-on hacking tools to steal sensitive data (2026-10-13)
- Chinese Government-linked Cyber Threat Actors Combine Automated and Hands-on Hacking Tools to Steal Sensitive Data (PDF) — FBI, CISA, NSA and NCSC-UK, Product ID AA26-281A, 2026-10-08, TLP: CLEAR (PDF, 58 pp.). The full advisory behindFootnote 11.
- Joint guidance on defending against the People's Republic of China-linked covert networks (2026-04)
- Joint guidance on the careful adoption of agentic artificial intelligence services
- Five Eyes cyber security agencies statement on the AI shift in cyber risk: why leaders must act now
- Best practices for event logging and threat detection
Public reporting
- Microsoft Security Research and Microsoft Threat Intelligence, AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise, 2026-10-21
- Wiz Research, ARTEX in the Wild: Storm-0940 Expands Targeting of IT and Software Vendors, 2026-10-22
- CrowdStrike, Unknown threat actor uses ARTEX to target South Korean finance
- Reuters, Chinese developer makes ARTEX AI agent closed-source after Korean bank hack, 2026-10-09
Advice and guidance
- Top 10 IT security actions to protect Internet connected networks and information (ITSM.10.089)
- Frontier artificial intelligence (ITSAP.10.050)
- Isolating web-facing applications (ITSAP.10.099)
- Network security logging and monitoring (ITSAP.80.085)
- Guidance on defence in depth for cloud-based services (ITSP.50.104)
- A zero trust approach to security architecture (ITSM.10.008)
- Developing your incident response plan (ITSAP.40.003)
- Ransomware playbook (ITSM.00.099)
- Social engineering (ITSAP.00.166)