A (0:00)

Cybersecurity Today would like to thank Meter for their support in bringing you this podcast. Meter delivers a complete networking stack, wired, wireless and cellular in one integrated solution that's built for performance and scale. You can find them at meter.com CST.

B (0:18)

Staples resells laptops in Canada with customer data still inside. A single click turns Microsoft Copilot into an attacker's accomplice. A critical AI flaw in ServiceNow and a highly advanced new Linux malware targets the cloud. And it's not your typical threat. This is Cybersecurity Today and I'm your host David Shipley. Let's get started. Canada's privacy commissioner is once again calling out a familiar personal data left behind on devices that get resold. A recent investigation found that Staples Canada resold returned laptops that contained customers personal information data that should have been wiped before those devices ever went onto shelves. And here's the part of this that really stings. It's not the first time Staples was flagged for the same issue roughly 15 years ago. Different decade, same failure. That points to an operational problem, not a technical one. Now, while this is embarrassing for a national retailer, let's be clear about all the consequences. Canada's privacy laws lag behind much of the world. They don't really have much teeth. Under that law called Pepita, the privacy commissioner can investigate and publicly report findings. But there are no meaningful fines, no GDPR style, multimillion dollar or higher penalties, no revenue based consequences. So the regulatory impact of this is fairly limited. But the secondary impact potentially is not. At a time when governments and industry are trying to encourage people to recycle electronics devices that can contain genuinely toxic materials. Stories like this don't just affect Staples reputation, they affect how people feel about recycling computing devices. If consumers believe recycling their laptop might mean their personal data ends up in someone else's hands, some will choose the worst alternatives, hoarding old devices or throwing them into the landfill. Staples says it's updating its practices. That's good news. But when the same issues resurface 15 years later, it's a reminder that privacy failures don't just erode trust in companies, they erode trust in systems we actually need people to use, like recycling old devices. Next up, a new class of attack researchers are calling Reprompt Targeting Microsoft Copilot, and it highlights a problem we already understand but still haven't fixed. Researchers showed that an attacker could hijack an active copilot session with a single user. Click a carefully crafted link injects hidden instructions into A channel copilot is designed to trust, and the AI executes those instructions using the victim's authenticated session. Here's the uncomfortable part. In some cases, researchers were able to bypass psychological copilot's guardrails simply by asking the same question twice. No exploit chain, no technical gymnastics, just persistence that tells you this isn't just a typical bug. It's a fundamental design problem with these large language models. Large language models don't reliably distinguish between content and instructions, and guardrails are often added after the fact. But that approach quickly turns into whack a mole. Every patch teaches attackers what to try next. And we've seen this movie before. Signature based endpoint security work Traditional antivirus until attackers learned to make small changes to defeat signatures. And what actually worked was an architectural shift systems that looked for changes in behavior and thought about whether something should or should not be happening. Sort of the behavioral based model of modern endpoint detection and reaction. And what's striking is that many of the companies selling machine intelligence today don't seem eager to apply that same approach when it comes to safety. When accuracy and error rates of their products affect revenue, they're relentless in trying to stamp out the problems. When abuse and security are the issue, we're told to wait. This problem will keep coming back until it hits vendors in the pocketbook. The physician won't heal itself, not until we demand it. Not until there's real consequences. Our third story involves ServiceNow, the enterprise workflow platform widely used for IT HR and security operations. Researchers recently disclosed a serious vulnerability affecting ServiceNow's virtual agent. In some configurations, attackers could impersonate legitimate users with minimal verification, in certain cases little more than an email address, and then interact with automated workflows as if they were authorized. Passwords and robust multi factor authentication were not always enforced. That authentication gap is the central issue. While the vulnerability involved AI powered agents, the underlying failure wasn't Novel systems, whether human driven or automated, can only be trusted to the extent the identity is properly verified. When authentication is weak, any process layered on top of it becomes so much easier to abuse. What automation changes here and AI agents do, is they amp up the scale. Once an AI enabled agent is acting on behalf of a user, actions that might normally require multiple manual steps can be executed quickly and consistently. ServiceNow has now patched the issue and says it has not observed active exploitation in the wild. The broader takeaway here is simple Getting identity and access management right for both humans and machines is even more important today, in the age of AI. Our final story today involves a newly discovered Linux malware framework that researchers describe as far more advanced than what we typically have seen. This isn't a single purpose payload. What it is is a modular platform with more than 37 modules designed for stealth reconnaissance and long term persistence. Before taking overt action, it gathers extensive information about its environment and what really stands out is its cloud first design. The malware can identify which major cloud provider it's running on and adjust its behavior accordingly, suggesting it was built to operate across aws, Azure, Google Cloud and others. Cloud providers often point out correctly that they have strong security architectures, but the cloud is also notorious for customer mistakes. Publicly exposed storage buckets and over permissioned identities remain common, and attackers only need a few missteps. In that context, a sophisticated Linux framework like this doesn't need to break the cloud, it just needs to wait for mistakes. Ironically, a small, tightly managed standalone Linux environment might actually be higher to compromise these days than some things that are hosted on these massive cloud platforms serving tens of thousands of customers. There is no evidence yet of widespread use of this new Linux malware in the wild, but the design signals a maturing threat model, one that assumes cloud infrastructure, Linux workloads and shared responsibility failures as the default environment before we wrap up, there's one common thread across all four stories. Whether it's laptops being resold with personal data still on them, AI assistants blindly following malicious prompts, enterprise agents acting without strong identity checks, or advanced malware targeting Linux in the cloud, none of these are zero days. None of these are surprises. They're the predictable result of convenience and complacency, outrunning control. Security failures today aren't happening because we don't know better. The failures here aren't just about technology. They go back to people, process and culture failures. They happen because of the basics. And if we want to have a better future, that has to change. That's it for Cybersecurity today. I'm David Shipley. Stay safe, stay informed and I'll be back on the news desk on Monday. Thanks for listening. We appreciate your support. If you like the show, please consider leaving us a like subscribing or even a review. And we need your help. Tell others about the show so we can reach even more people in 2026. Thanks so much. Have a great weekend.