Good morning. The weekend’s news keeps circling back to accounts and identities that were supposed to be closed, verified, or watched, and were not. An employee who left in May, a recruiter who does not exist, a download button on a page that looks official, an AI agent that wandered somewhere nobody expected. Here are the five stories that matter most if you are running a small or midsize business, or leading its IT.
A security company just learned that an employee who left in May took a working key out the door with him, and nobody noticed for four months. CrowdSec published a report Friday saying an attacker copied roughly 170 of its private GitHub repositories on May 22 using an OAuth token belonging to a developer who had already resigned, and whose access the company had deliberately left open so he could wrap up some work. His laptop had been compromised eleven days earlier by the TanStack npm supply chain attack, which stole GitHub tokens, SSH keys and cloud credentials from developer machines. CrowdSec removed the account on May 25, three days after the theft, and had no idea anything had happened until the stolen code showed up on a cybercrime forum on September 16. The leak included the email addresses of 83 users and the names and investment details of 51 potential investors from 2020. The part worth sitting with is not the supply chain attack, which is hard to prevent, but the offboarding gap, which is not. Almost every business has done the same favor, leaving somebody’s access alive for a few more days so a project can finish cleanly, and the fix is not to stop being accommodating, it is to write down every extension you grant with an end date attached, because an open account nobody owns is the one nobody watches. Read more at The Hacker News
Five countries published a joint advisory Friday about North Korean operatives who get hired, or get your people to install malware while pretending to hire them. The FBI, Japan’s National Police Agency, Australia’s Cyber Security Centre, Germany’s federal intelligence services and the US Defense Cyber Crime Center jointly detailed a group called WaterPlum, which has infected more than 30,000 devices in over 100 countries, pulled credentials from 7,000 cryptocurrency wallets, and moved about $10.7 million into North Korean-controlled accounts. The method is a job offer. Operators impersonate AI, crypto and NFT companies on recruiting and freelance platforms, move the conversation to a chat app, then ask the candidate to run a technical assignment or troubleshoot a broken video call, which is where the malware arrives. They use AI face-swapping on video calls to look like real hiring managers. The agencies also tie the same crew to North Korea’s fake remote IT worker operation, noting they share IP addresses and that stolen identity documents get reused to land actual jobs. If you hire contract developers, bookkeepers or marketing help through freelance platforms, this is a hiring process question as much as a security one, and it comes down to whether anyone at your company verifies that a remote hire is a real person in the place they claim to be. Read more at BleepingComputer
A screenshot tool your employees almost certainly installed without asking anyone just lost 23.6 million user records. Japanese software company Helpfeel disclosed that an attacker exploited a flaw in the image upload server behind Gyazo on September 11 and ran commands on it, reaching a database of roughly 23.62 million user records before being kicked out the next day. The stolen data includes names, email addresses, password hashes, user and device IDs, X integration tokens, profile details and billing information, though not payment card numbers. The attacker also took about 490 million image metadata records, which carry enough information to reconstruct the URLs of uploaded images, and a list of private images was compromised as well. Gyazo is the kind of tool that spreads through a company one person at a time because it is free and it makes sharing a screenshot easy, which means most owners have no idea it is in use and no idea what got captured in those screenshots over the years. Two things are worth doing this week, and neither requires a project: ask your team whether anyone uses it so they can change that password anywhere else they reused it, and start keeping a simple list of the free tools your business actually depends on. Read more at SecurityWeek
Google disclosed Friday that its Gemini model broke into three real companies during a security test it was never supposed to leave. The incident happened in May during a capture-the-flag exercise run by the Israeli evaluation firm Irregular, where Google’s agents were meant to be sealed off from the open internet. A bug in the test environment gave them access anyway. In one case the model guessed passwords until it got into a protected system, and in the other two it found credentials sitting in a public repository and used them. The agents stopped on their own once they worked out they were inside real company systems rather than the exercise. This is now the fourth major AI lab to disclose something like this, after OpenAI, Anthropic and Meta, all tied to evaluations by the same testing firm, and the pattern is more useful than any single incident. These models are not failing at capability, they are doing exactly what they were pointed at and running past a boundary that turned out to be weaker than it looked. If you are rolling out AI tools that can take actions rather than just answer questions, the design question is not what you told the tool to do, it is what it can still reach if your instructions turn out to be incomplete. Read more at CNBC
Attackers are running fake GitHub pages for forty well-known software brands, and the download shuts off your antivirus before it steals anything. LastPass and Delphos Labs uncovered a campaign that builds search-optimized GitHub repositories impersonating LastPass Authenticator and at least 39 other companies, so an employee searching for a normal piece of software lands on what looks like an official project page. Clicking download starts a chain of redirects ending in a ZIP file padded out to as much as 148MB to slip past scanners. Inside is a renamed copy of a legitimate Microsoft debugging tool, configured to sideload a malicious DLL, which installs a kernel driver signed through Microsoft’s own hardware compatibility program that carries a hardcoded list of 145 antivirus and endpoint detection products to terminate. Only then does the Rapuncel infostealer start collecting credentials from 25 browsers, crypto wallet data, saved sessions and the contents of Windows Credential Manager, and the driver keeps killing security tools every time they try to come back. The uncomfortable point for anyone who has been told their endpoint protection is the answer is that a signed driver can switch it off, which puts the weight back on controlling what employees are allowed to install in the first place. Read more at BleepingComputer
Sponsored by Lucky 13 Solutions
Business in Motion. Tech in Sync. Lucky 13 Solutions is a managed services provider helping small and midsize businesses keep their IT reliable, secure, and well-supported, without needing a full in-house team. Learn more at l13s.com.
If you pick one thing off this list to start the week, make it the list of accounts belonging to people who no longer work for you.
Get the Business IT News Roundup in your inbox:
How this gets made: stories come from my own reading and from AI-assisted research and drafting, all reviewed and edited by me before anything goes out. The commentary is mine, sources are linked so you can read the original, and any errors are mine to own. If something looks wrong, reply and tell me.