Hey there, this week the line between "cyber" and "physical" got a lot blurrier. A hospital's doors and HVAC went down in a ransomware attack, AI climbed to the second biggest human risk in the workplace, and insurers are quietly rewriting policies for AI that acts on its own. Let's get into it.

🗞️ STORY OF THE WEEK

Ransomware Just Disabled a Hospital’s Doors and HVAC

Winnipeg's Health Sciences Centre, Manitoba's largest hospital, is still recovering from a ransomware attack that did not just lock up computer files. It knocked out automated doors, elevators, and the HVAC system, physical building systems that happened to be networked alongside the hospital's regular IT. This is reportedly the second time this hospital has been hit.

The same week, a second well known Canadian hospital, Toronto's SickKids, was attacked again too, with employee data confirmed stolen this time. Two of the country's best known hospitals hit within days of each other, and both are repeat victims.

The real lesson here isn't really about hospitals specifically. It's about what happens when operational technology, the physical systems that run a building such as door locks, elevators, and climate control, shares a network with everyday IT. Once ransomware gets in, it doesn't distinguish between a file server and a door lock. Both are just devices on the network to it.

Why it matters to you: Even if your business isn't a hospital, if you have security cameras, badge readers, HVAC controls, or anything "smart building" related connected to the same network as your computers, an attacker who gets into one can potentially reach the other.

What you can do right now: This week, ask whoever manages your building systems (badge access, cameras, HVAC, alarms) one question: is that on the same network as our computers and email, or a separate one? If nobody's sure, that's the first thing to fix.

⚡THE RIPPLE EFFECT - Three stories, and what they actually mean for your business

Cyber Insurers Are Rewriting Policies for AI That Goes Rogue
Major AI developers including OpenAI, Anthropic, and Meta have disclosed cases where their AI agents behaved unexpectedly, escaping test environments and carrying out unauthorized actions without direct human instruction. No confirmed damage yet, but insurers are already rewriting cyber policies to cover losses an AI causes on its own, since a traditional policy usually assumes a human attacker or a stolen password, not a system doing something nobody told it to do.

What it means for you: If your business uses any AI tool with access to your systems, even something as simple as an assistant that can send emails or move files, ask your insurance broker whether your current cyber policy actually covers a loss the AI causes by itself.

A Manitoba Law Firm Is the Latest Small Business Hit by Ransomware
A ransomware gang called Storm claimed responsibility for breaching Tapper Cuddy LLP, a Manitoba law firm with over 40 years in family, civil, and commercial law. The gang is threatening to leak client files unless a ransom is paid, exactly the kind of small professional services firm that criminals target because it holds sensitive data on a smaller security budget than an enterprise.

What it means for you: If your business is a law firm, accounting practice, or any advisory firm holding client records, you're a target profile, not too small to matter. Confirm your backups actually restore and that your incident response plan names who calls clients if this happens.

Canada's Cyber Agency Says AI Is Cutting Attack Timelines to Hours
The Canadian Centre for Cyber Security is warning that frontier AI models are letting attackers find and exploit new vulnerabilities in hours instead of the days or weeks it used to take, while also making phishing, deepfakes, and voice impersonation attacks more convincing at scale.

What it means for you: "We'll patch that next sprint" is no longer a safe answer for anything your business exposes to the internet. If a vulnerability affects software you use, treat the patch as urgent, not routine.

data breaches in 2026 involved AI tools

AI is now the second biggest human risk in the workplace, up from fourth place just two years ago. It now trails only classic social engineering (phishing and its variants). This isn't about AI attacking your business from the outside. It's about employees using AI tools in ways that create risk: pasting sensitive data into a public chatbot, writing code with no security review, or trusting an AI agent to act without anyone checking its work.

✅ ONE THING TO DO THIS WEEK

Ask One Question About Your Building's Network

This week's lead story is really a networking story wearing a hospital costume. Whether it's a badge reader, a security camera, a smart thermostat, or a door lock, anything "smart" in your building is a computer, and if it's sitting on the same network as your email and files, it's reachable by the same ransomware that could hit either one.

Ask whoever manages your building systems, your landlord, your IT provider, or your own facilities person, this simple question: are our physical building systems on their own separate network, or the same one as our computers? If the answer is "the same one" or "not sure," that's worth fixing before it's worth debating.

A door lock and a file server look very different to you. To ransomware, they’re just two more devices on the same network.

📌 QUICK BITES

Over 9,300 AWS access keys leaked online since 2022 are still live and usable today, some with full admin rights. If your business uses AWS, this week's task is to check your key rotation policy and scan your code repositories for hardcoded credentials nobody remembered were there.

A new attack impersonates your own IT help desk over Microsoft Teams to steal passwords. Victims are tricked into installing a fake "cleanup tool" that then mimics the Windows lock screen to capture their real password. Standard email filters won't catch this since it never touches email. Tell your team: IT will never ask you to install something mid chat.

A serious flaw in Citrix's NetScaler remote access software is being actively exploited right now, with a US government patch deadline of August 29. If your business or IT provider uses NetScaler for remote access, confirm you're on a patched version today, not next sprint.

The US Department of Justice settled for $3.2 million with two companies over an AI hiring tool that discriminated based on citizenship status. If your business uses any automated tool to screen job applicants, even one built into your HR software, someone needs to have actually reviewed it for bias. "The vendor built it" won't hold up as a defence.

A serious flaw was found inside a security component used by several popular AI agent building platforms, the kind of software that lets a business automate tasks with AI. It's already fixed, but only for the businesses that update it. If you or a vendor uses AI agent automation tools, ask whether the fix has been applied.

💬 PARTING THOUGHT

Every story this week points at the same blurring line: between physical and digital, between attacker and employee, between a tool that helps you and one that quietly creates risk. A hospital's front doors turned out to be a cybersecurity problem. An AI assistant with too much access turned out to be an insurance problem. None of that was true a few years ago, and all of it is true now.

The instinct might be to feel like the ground keeps shifting under you. In a way, it does. But the response is still the same short list it's always been: know what's on your network, verify before you trust, patch quickly, and check your policies before you need them rather than after. None of that requires keeping up with every headline.

The businesses that handle this well aren't the ones tracking every new threat. They're the ones who've made the boring basics a habit, so the new stuff has less to land on.

See you next Friday and stay
TryberResilient