Canada’s cybersecurity agency says attackers are now handing much of their work to AI agents, and it’s shrinking the time defenders have to react. The Canadian Centre for Cyber Security, part of the Communications Security Establishment, published a threat bulletin this week on what it calls agentic AI accelerating cyberattacks.

What it found

The bulletin centres on a group Microsoft tracks as Storm-0940, which used an AI agent framework called ARTEX against hundreds of organizations. The framework strung together several stages of an attack: choosing targets, testing passwords and exploring systems once inside.

  • In some cases, the group went from selecting a target to privileged access to cloud systems in less than three hours.
  • Targeted sectors include financial services, manufacturing, defence, automotive, media and technology, The Globe and Mail reports.
  • The attacks aren’t fully autonomous. People still pick targets, set goals and supervise. But more of the work in between can be delegated to machines.

The agency’s main worry isn’t new hacking techniques. It’s that AI links, automates and repeats existing ones, creating what the bulletin calls a “reusable framework that can help conduct cyber operations on an industrial scale.” It expects this kind of activity to become more targeted and harder to spot.

“We do see AI enhancing and being used across different elements in the kill chain,” Rajiv Gupta, head of the Cyber Centre, told the Globe.

What it recommends

None of the advice is exotic. For organizations, the Cyber Centre says to:

  • require phishing-resistant multi-factor authentication for administrators, remote access and other important accounts
  • limit what users, apps and service accounts can access, and review access granted to third-party cloud apps
  • patch internet-facing systems promptly and turn off services you don’t need
  • keep security logs and watch for unusual sign-ins and new credentials
  • decide in advance who’s allowed to cut off accounts or networks, and practise responding to incidents that unfold within hours
  • keep tested offline backups

It adds extra steps for critical infrastructure operators, including testing manual ways to keep running if systems have to be isolated.

Why it matters: For years, the gap between a breach and its discovery gave defenders some time. If an AI-assisted attacker can reach cloud admin access in an afternoon, that time is gone. The fixes are the familiar ones, MFA and patching, but the Cyber Centre’s message is that “we’ll get to it” is no longer good enough. For individuals, the same lesson applies: turn on passkeys or app-based two-factor wherever you can.