This week the call came from inside the building, or at least it sounded that way. Two of Canada's best known law firms were breached after attackers apparently posed as their own IT staff, and the Canadian Cyber Centre says that tactic is being aimed at Canadian organizations right now. Meanwhile RBC has told AI assistants to stay out of its online banking, following last week's bank story, and Binance lost an estimated $6.3 billion after a single engineer's laptop was compromised. Different targets, same weakness: a trusted-looking request. Let's get into it.
🗞️ STORY OF THE WEEK
Two Top Canadian Law Firms Were Breached, and the Attackers Pretended to Be IT
McCarthy Tétrault, one of Canada's largest corporate law firms, confirmed this week that an unauthorized party accessed some of its files, including information relating to clients. The firm says it found the incident late on October 22, notified affected clients on October 25, and described it as a coordinated social engineering campaign. Another Bay Street firm, Davies Ward Phillips & Vineberg, reported a breach that compromised information from one account. Both say they have notified law enforcement.
Details are still emerging and some come from unnamed sources. The Globe and Mail reported, citing a source, that roughly 31,000 files were taken from McCarthy Tétrault, a figure the firm has not confirmed, and that in some cases the attackers used Microsoft Teams, posing as internal IT staff, to ask an employee for access to their computer. That fits a warning the Canadian Centre for Cyber Security issued on October 28: an ongoing voice phishing campaign in which attackers pretend to be a company's help desk, steer employees to fake sign-in pages, capture passwords and one-time codes as they are typed, then register their own devices for multi factor authentication.
The legal sector is being singled out, but nothing about the technique is specific to lawyers. LawPro, the insurer for Ontario law firms, has said recent cyberattacks on firms have led to unauthorized withdrawals totalling millions of dollars. Any business that holds client money, contracts or personal records, and has an IT helper people trust, is a possible target. The attacker needs no software flaw, only one convincing phone call or message.
Why It Matters
Multi factor authentication is not enough if an employee reads the code out loud to someone who sounds like IT. Ask yourself whether your staff know what a real request from your IT provider looks like, and what to do when one feels slightly off.
What You Can Do
Tell everyone this week: if someone calls, chats or messages claiming to be IT and asks for a password, a code, or remote access to your computer, hang up and call IT back on a number you already know. Ask your IT provider to confirm that resetting a person's multi factor authentication needs a second verification step.
⚡THE RIPPLE EFFECT - Three stories, and what they actually mean for your business
RBC Tells AI Assistants to Stay Out of Its Online Banking
Last week we covered US banks finding AI agents inside customer accounts. This week it reached Canada. A memo sent to RBC staff on October 27, confirmed by the bank, bars third party personal AI agents from its authenticated digital services, including online banking. RBC cited fraud, security and privacy risk, noting that agents need logins and may store passwords. Existing agent connections will stop working, though customers can still bank through RBC's website and app. Reports say TD is taking a similar step and other Canadian banks are limiting agent access, while Canada's banking regulator says it is increasing its oversight of AI at the big banks.
What it means for you: If you or your staff have connected an AI assistant to any bank or payment account, expect it to stop working and expect your other financial institutions to follow. Make sure nothing in your month end process quietly depends on one.
Binance Loses an Estimated $6.3 Billion in the Largest Crypto Theft Ever
Binance, the world's largest crypto exchange, paused withdrawals after attackers moved an estimated $6.3 billion out of its wallets, mostly within about 20 minutes. Binance says the breach began with a compromised laptop belonging to one engineer, which gave the attacker access to a service that could submit withdrawals. Blockchain analytics firm TRM Labs attributes the theft to North Korea's TraderTraitor group, while some outlets say attribution is not settled. Binance has said it will cover losses, and reports say billions in assets have been frozen.
What it means for you: You do not hold billions, but the pattern is the same as the law firms: one person's device or account becomes the way in. Ask your IT provider who has administrator access in your business, and whether those people use hardened, company managed devices for it.
The Crime Stoppers Breach Is Triggering a Federal Privacy Investigation
The Privacy Commissioner of Canada opened an investigation into a reportedly large breach of the Crime Stoppers tip platform, announced October 27. Police departments across the country paused accepting tips through the system while it is examined. The scope, data types and timing have not been disclosed, and CBC reporting describes it as potentially one of the most harmful breaches Canada has seen because of the sensitivity of tipster information.
What it means for you: Any outside platform that holds sensitive submissions for you, such as an HR hotline, whistleblower form or client intake portal, is part of your risk. Ask the vendor how the data is protected and how fast they would tell you about a breach.
32.5 million
Last week's lead story, the PHP2Shell flaw, moved from warning to confirmed attacks. The US cyber agency CISA added both flaws to its list of actively exploited vulnerabilities on October 26. Internet scanning firm Censys estimates more than 32 million internet facing PHP services still run an affected version. That is exposure, not confirmed compromise, and security firms disagree on how many attacks have succeeded, with two vendors withdrawing earlier claims. The advice has not changed: confirm in writing that your host has installed the fix.
✅ ONE THING TO DO THIS WEEK
Add a Rogue AI Agent Scenario to Your Incident Plan
The US agencies NIST, CISA and the NSA published a ten practice baseline this week for responding to incidents involving rogue AI agents, and the industry attack catalogue MITRE ATT&CK now lists prompt injection and AI agent abuse as formal threats. This week's news shows why: RBC is shutting agents out, and OpenAI disclosed its own agents have escaped virtual machines.
You do not need a new plan. Open the incident response plan you already have, or write one page if you do not, and add a single scenario: an AI tool or assistant connected to our systems does something we did not ask it to. Write down who can switch it off, how you would find out what it touched, and who you would call. Then run through it once with your team in fifteen minutes.
If you can’t switch an AI tool off quickly, you haven’t finished setting it up.
📌 QUICK BITES
A flaw in a widely used web building block called Brotli is being actively exploited, and the Canadian Cyber Centre has issued an alert. It affects software built with Node.js, among other tools, and the US cyber agency has added it to its must patch list. Ask your developer or hosting provider whether your site or app is affected, and whether the software version is still supported.
Loblaw signed a privacy compliance agreement with the Privacy Commissioner after a February breach affected about two million Canadians. It committed to an independent security assessment and to report how it will act on the findings. Expect regulators to want proof of fixes after a breach, not just an apology.
IBM announced identity controls for AI agents, including individual identities, limits on what they can reach and fast permission revocation. The idea is simple and worth copying: treat each AI agent like an employee with its own login, minimum access and an off switch.
Researchers found a booby trapped instruction file in a customer's Downloads folder that tried to trick an AI coding assistant into installing malicious software. The attack was stopped before anything ran. If your developers use AI coding tools, require a person to approve any software they install.
OpenAI says its agents have escaped virtual machines and communicated across sandboxes, and it is training new models without live internet access. Containment has to be built in from the outside. If an AI tool touches your systems, ask what limits and logs exist around it.
💬 PARTING THOUGHT
Look back over this week and almost nothing required a clever exploit. A phone call that sounded like IT. A laptop that belonged to one engineer. A login handed to a bot. Each was a request that looked reasonable and wasn't.
That is oddly reassuring, because the defence is not exotic either. It is a pause before acting on a request, a second way to verify who is asking, and a habit of knowing who and what holds the keys.
Resilience is mostly the discipline to hang up and call back.
See you next Friday and stay
TryberResilient