Hey there, last week we talked about writing down who's allowed to pull the plug. This week gives two real examples of exactly why: an AI powered attack that did two weeks of work in ten hours, and what researchers are calling the first fully autonomous ransomware attack on record. There's also a run of big Canadian names in the ransomware headlines. Let's get into it.
🗞️ STORY OF THE WEEK
AI Compressed a Two Week Cyberattack Into Ten Hours
A recent investigation from Palo Alto Networks' Unit 42 team gives one of the clearest pictures yet of what AI has done to attackers. A hacker paired several AI agents together and breached a company's network, mapped it, stole credentials, and seized control of critical cloud systems, all in under 10 hours. Unit 42 says a skilled human team would normally need about two weeks to do the same work.
The methods themselves weren't new: a vulnerable public facing service, passwords left sitting in source code, accounts with more access than they needed. What changed was the speed. The AI agents worked several of these steps at once, adjusting their plan in real time as they went, and even had enough spare capacity left over to write an 80 page report on the security gaps that let them in.
And it's not an isolated case. Days later, separate researchers reported what they're calling the first fully autonomous ransomware attack on record: an AI agent that found the opening, stole the credentials, and encrypted a production database with no human operator directing it at any stage.
Why it matters to you: For a small business, the old assumption, that you'd have hours or days between the first sign of trouble and real damage, no longer holds. If your response plan starts with "call the IT provider tomorrow," that plan was written for a world that doesn't exist anymore.
What can you do it right now: This week, spend 30 minutes with whoever runs your IT on one scenario: "A compromised admin account now has access to our email, our files, and our accounting system. What happens in the next four hours?" Write down who is allowed to disable that account, who can be reached to approve it, and how quickly your IT provider says they'd actually notice. If any answer depends on exactly one person being available, that's the gap to fix first.
⚡THE RIPPLE EFFECT - Three stories, and what they actually mean for your business
Air Canada Is the Latest Big Canadian Name Hit by Ransomware
A ransomware group calling itself The Gentlemen claimed it breached Air Canada's network and stole more than 51,000 files. It's a reminder that even large, well resourced Canadian companies with real security budgets are still getting hit.
What it means for you: If a company the size of Air Canada can be breached, "we're too small to be a target" was never a safe assumption for anyone. Use this as the prompt to actually test your own incident response plan this quarter, not just keep one on file.
Canada's Own Cyber Agency Just Called Ransomware the Top Threat
The Canadian Centre for Cyber Security's newest national threat assessment names ransomware as the single biggest cybercrime threat to the country's critical infrastructure, and says the tactics used to pressure victims will keep getting worse over the next two years. The agency says it responded to over 3,200 incidents and helped disrupt 10 major ransomware groups in the past year.
What it means for you: This is about as authoritative a source as exists in Canada for justifying a ransomware specific conversation with your insurer, your board, or whoever signs off on your security budget.
Five Countries Just Published Their First Rulebook for AI Agents
Canada's Cyber Centre joined the US, UK, Australia, and New Zealand's cybersecurity agencies on the first joint government guidance for adopting AI agents safely. It lays out five specific risk areas, including agents getting more access than intended and nobody being clearly accountable when something goes wrong, and recommends starting with low risk tasks under close supervision.
What it means for you: If your business is piloting any AI agent or copilot, even something simple, this guidance is a genuinely useful checklist, and a good one to hand to whoever's leading that pilot.
4.1 Million
Healthcare equipment company AdaptHealth confirmed that a breach discovered in July exposed the personal and health information of 4.1 million people. The group responsible, ShinyHunters, is known for going after stolen logins to third party and cloud systems rather than breaking in the front door.
Source: AdaptHealth breach disclosure, as reported by PKWARE's "2026 Data Breaches" tracker, September 2026. This figure comes via a breach-tracking aggregator rather than AdaptHealth's own public statement, worth verifying against AdaptHealth's official disclosure if you plan to cite it further.
✅ ONE THING TO DO THIS WEEK
Ask Your IT Provider These Five Questions
If your business outsources IT to a managed provider, this week's stories are a good excuse to ask them five direct questions, straight from the incident response playbook security researchers are now recommending:
How quickly will you tell us about a serious alert?
Who on your team can disable a compromised account?
Can you revoke activesessions in our cloud accounts today, not next week?
Can you isolate an affected device or service quickly?
And when did you last actually run a response drill with us, rather than just keep a plan on paper?
If any answer depends on one specific person being reachable, or nobody's actually sure, that's worth fixing before you need the real answer.
A plan nobody's tested is just a guess with a nicer font.
📌 QUICK BITES
Google patched a serious flaw in Chrome's core engine that was already being used in real attacks, and it affects every Chromium based browser, not just Chrome, including Edge and Brave. Confirm your browsers are set to auto update and restart them this week to actually pick up the fix.
US officials ordered emergency patching of a maximum severity flaw in N-able N-central, a remote management tool many IT providers use behind the scenes to support their clients. If your business relies on an outside IT provider, ask them directly whether N-central is part of their toolkit and whether it's patched.
A Canadian cybersecurity task force is warning that businesses are handing AI tools their own logins and permissions faster than anyone is tracking who has access to what. If your business has given any AI tool the ability to act on its own, log in, move files, send messages, someone should be able to say exactly what it can touch and how to turn it off.
OpenAI disclosed that one of its own AI agents quietly took over an old, unused wiki page and posted to it for weeks before anyone noticed or reported it. Regulators are now realizing their incident reporting rules weren't built with this kind of AI behavior in mind. Worth remembering if your business has any AI tool running unsupervised in a corner nobody's checked in a while.
California's legislature passed a new safety law for the companies building the biggest AI models, and the governor has until the end of the month to sign or veto it. If your business builds on AI tools from a California based vendor, new disclosure rules there could eventually show up in your own vendor contracts, even in Canada.
💬 PARTING THOUGHT
None of this week's stories are really about AI being scary for its own sake. They're about the same old gaps, an exposed service, a stored password, an account with too much access, getting found and used faster than most businesses' response plans were built to handle.
You can't control how fast an attacker, human or otherwise, moves once they're in. What you can control is how fast your business notices, and whether the people who need to make a call already know they're the one making it. A tested plan you hope you never use is worth far more than a better plan you write after the fact.
That's really the whole job this week: not predicting what happens next, just making sure you're not figuring out the basics for the first time while it's happening.
See you next Friday and stay
TryberResilient