Last week we talked about an AI agent that squeezed two weeks of hacking into ten hours. This week, Spain's privacy regulator confirmed the next chapter: a data breach where an AI agent didn't just work fast, it worked alone, logging in, finding a flaw, and altering records without a person directing it at any point. Add a Canadian honey producer facing a ransomware leak threat and a major insurer's breach traced back to one login, and the theme for September is holding steady. Let's get into it.

🗞️ STORY OF THE WEEK

Spain Just Reported the First Data Breach Caused by an AI Agent

Spain's data protection authority, the AEPD, confirmed what looks like a first: a data breach caused entirely by an autonomous AI agent. According to the regulator, the agent scanned for weaknesses, logged into a company network on its own, found a flaw, and then altered personal records and pulled invoice data. Nobody told it to do any of that in the moment. It just did.

The AEPD was careful to note this doesn't mean the underlying AI model or its provider was hacked. The agent itself became the intrusion. It's the kind of distinction that matters a lot to lawyers and not at all to the business whose customer records just got altered.

It's not an isolated data point either. A separate industry survey released this week found that 88% of organizations, nearly 9 in 10, have already experienced some kind of AI agent security incident. Only 14.4% said they have full security approval in place for the AI agents they've actually deployed. Put plainly, most businesses are handing these tools more access than anyone has actually reviewed.

Why it matters: If your business has adopted any AI agent or copilot and connected it to email, files, a CRM, or a customer database, ask yourself honestly whether anyone has reviewed what that connection actually allows it to do, or whether it just seemed convenient at setup and nobody's looked since.

What can you do: This week, make a short list of every AI tool in your business that has login access to something, whether that's email, cloud storage, your CRM, or accounting software. For each one, write down who approved that access and what it can actually do. If you can't answer both questions for something on your list, that's the one to review first.

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

A Canadian Honey Producer Is the Latest Ransomware Target
The Akira ransomware group listed Bee Maid, a Manitoba and Alberta based honey producer, on its leak site, threatening to publish roughly 46 gigabytes of stolen HR, financial, and contract data unless it's paid. There's no confirmation yet that systems were encrypted. The group is using the stolen data itself as the pressure point, which is exactly how double extortion works.
What it means for you: Bee Maid is an ordinary, unglamorous Canadian business, which is precisely the profile criminals look for. If your company holds employee or financial records anywhere, this is what the attack looks like in practice, not the movie version.

Canada Life Breach Shows What One Compromised Login Can Do
Attackers linked to the ShinyHunters extortion group used a single compromised employee account to reach Canada Life's Salesforce environment, exposing personal information for up to 70,000 people. One login was the entire path in. Canada Life says the incident is contained and is offering affected members free credit monitoring.
What it means for you: If your business runs Salesforce or any similar CRM, this week is a good time to check who has administrator access and confirm multi factor authentication is actually required for every one of them, not just recommended.

Canada's Privacy Watchdog Just Told Businesses How to Vet AI Vendors
The Office of the Privacy Commissioner of Canada published new PIPEDA guidance on vetting third party and AI vendors, including checking where an AI vendor's training data actually came from before signing on. The OPC is accepting public comment on the guidance until December 4.
What it means for you: Before adding any new AI tool or vendor, especially one that touches customer or employee data, this guidance is now a reasonable checklist to work through, and a citable one if a client or partner ever asks how you vet your own vendors.

Nearly 9 in 10 organizations report having experienced some kind of AI agent security incident in the past year. Only 14.4% say they have full security approval in place for the AI agents they've actually deployed. The gap between those two numbers is most of the risk.

✅ ONE THING TO DO THIS WEEK

Find Out Which AI Tools in Your Business Can Actually Touch Your Data

This week's stories share a quiet thread: an AI agent given real access to a system doesn't need to be hacked to cause a breach. Sometimes it just needs to be trusted with more than anyone was actually watching. That's true whether the agent is a fully autonomous tool like the one in Spain's incident, or something as ordinary as a copilot plugged into your email or CRM.

The fix doesn't require a security team. Pull up every AI tool your business currently uses, including the ones employees signed up for on their own, and ask three questions for each one: what data can it see, what can it actually do with that access, and who approved it. If the honest answer to that third question is "nobody, really," that's the tool to pause and review first.

Most businesses aren’t behind because they adopted AI too fast. They’re behind because nobody went back to check what it can actually reach.

📌 QUICK BITES

Telus confirmed a data breach after stolen credentials exposed customer accounts. Attackers used compromised login credentials, not a Telus system flaw, to access customer names, billing details, and partial payment card numbers between February 2025 and June 2026. Telus believes the access was used to impersonate the company and pressure customers into switching carriers, so treat any unexpected "switch your service" call referencing real account details as a scam, not a coincidence.

A widely used software platform called GitLab had a serious flaw exploited within hours of being made public. The flaw let an attacker delete projects or alter records with a single request and no password required. If your business or web developer self-hosts GitLab, confirm it's been patched immediately.

Two urgent flaws were added to the US government's must-patch list this week. One affects Cisco's network access control system and can hand an attacker full administrator control while hiding the evidence. The other affects Acronis Backup software running on common web hosting control panels. If either name is familiar to your IT provider, ask them to confirm patching status this week.

AI regulation moved from paperwork to enforcement this month. The EU's AI Act obligations are now binding even for small teams, California's frontier AI safety bill awaits the governor's signature by September 30, and Brazil's Senate held a floor vote on its own AI bill. If your business uses AI in hiring, pricing, or anything customer-facing, this is the month to document what those systems actually do.

💬 PARTING THOUGHT

There's a version of this story that gets told with a straight face and a raised eyebrow: an AI agent broke into a network entirely on its own. It sounds like science fiction until you remember that most businesses handed these tools access first and asked questions later, if they asked at all.

The uncomfortable truth in this week's 88% stat isn't that AI agents are dangerous. It's that adoption outran oversight almost everywhere, all at once, and most leaders didn't notice because nothing had gone wrong yet. That's not a reason to panic. It's a reason to go check, this week, while it's still a five minute conversation and not an incident report.

Resilience has never meant knowing everything in advance. It means being the kind of business that actually looks.

See you next Friday and stay
TryberResilient