Good afternoon. Tuesday’s news kept coming back to the same quiet problem, which is how often the thing that puts a business at risk is something nobody was told about. A plugin gets patched without an announcement, a vendor bug sits in the wild for months, an AI tool ships with a setting that made sense to an engineer and not to anyone else. Here are the five stories that matter most if you are running a small or midsize business, or leading its IT.
Attackers are hunting WordPress sites running a single sign-on plugin that got quietly patched without anyone being told it was a security fix. SecurityWeek reported Tuesday that threat actors are opportunistically exploiting CVE-2026-61979 and CVE-2026-15981 in the MiniOrange SAML 2.0 SSO plugin, a pair of authentication bypasses that let an attacker log in as any WordPress user including an administrator. All seven editions of the plugin were fixed, but only the free version got an advisory, and even that one described the change as a bugfix rather than a security patch, so paid customers had no reason to act and their dashboards will not prompt them to update. Patchstack put it well when it said the attacker does not need to know which edition you run, because you do, and that is the uncomfortable part for any business whose website is quietly running plugins nobody has looked at in a year. If your site uses MiniOrange for logins, someone needs to check the version by hand today, and if you cannot name who is responsible for keeping your website’s plugins current, that is the bigger finding here. Read more at SecurityWeek
A criminal phishing service is now running AI voice agents that call victims, pose as Apple support, and talk them out of their passcodes for about ten cents a call. BleepingComputer reported Tuesday on AnonyMousKIT, a phishing-as-a-service platform SOCRadar traced across 506 domains and 168 reseller storefronts, built to unlock stolen iPhones by harvesting Apple IDs and Activation Lock codes. Researchers recovered records of 200 calls placed to victims between August 2025 and May 2026, handled by a voice AI agent working from five different personas, one of which introduces itself as “Alice from Apple Support” and tells the victim their stolen phone turned up at an Apple store. The detail worth carrying into your own business is the price, because ten cents a call means voice social engineering is no longer reserved for high-value targets, and the same economics work just as well against your accounts payable clerk as against a stolen phone owner. SOCRadar also found some of these messages landing at corporate and government addresses, which is a reminder that a compromised personal Apple account can carry work email, saved passwords, and company files right along with it. Read more at BleepingComputer
Visiting one bad web page can permanently rewrite the instructions your local AI agent follows, and the agent will never know it happened. Cyera’s Oasis Identity Research disclosed a flaw in NVIDIA NemoClaw, the tool used to deploy the OpenClaw agent framework, in which the packaged Ollama model server is bound to all network interfaces instead of the loopback address and skips its browser protection check as a result. Dark Reading reported Tuesday that an attacker can use a malicious page and DNS rebinding to reach that server without any credentials, then modify the model’s chat template so hidden instructions get appended to every system prompt from that point forward, surviving across conversations and staying invisible to both the agent and the person using it. The fix is out for macOS and Linux in version 0.0.35, with no fix yet on Windows. What makes this worth your attention even if nobody at your company has heard of NemoClaw is the pattern, since the danger came from an ordinary infrastructure shortcut rather than anything exotic, and if your team is running AI agents with access to files, code, or internal systems, the question to ask is not whether the model is trustworthy but who else can reach it. Read more at Dark Reading
A maximum-severity Oracle flaw that was patched in January has been under attack all year and only just made the federal must-fix list. SC Media reported Tuesday that CISA added CVE-2026-21962 to its Known Exploited Vulnerabilities catalog on August 24 and gave federal agencies until August 27 to patch it, a three-day window that reflects how aggressively the flaw is being used. It affects Oracle HTTP Server and the WebLogic Server Proxy Plug-in, requires no credentials, and lets an attacker reach whatever data sits behind that proxy. Security researchers quoted in the piece noted that CloudSEK saw high-volume automated attacks within days of exploit code appearing on January 22, and that a China-linked group was already using it against government targets by July. The line worth writing down came from SafeBreach’s Adrian Culley, who said patched since January is not the same as watched since January, because these front-door components tend to sit outside the inventory and monitoring that the applications behind them receive. If your company runs any internet-facing middleware, the practical question is whether it appears on an asset list at all. Read more at SC Media
New CompTIA research says companies have stopped experimenting with AI and started trying to actually wire it into how the business runs, which is where it gets hard. The inaugural Corporate AI Adoption report, published Tuesday and based on a June survey of 1,027 business and technology professionals, found 59% of organizations now prioritize integrating AI into their existing systems and processes rather than simply putting more tools in employees’ hands. Ninety-three percent expect their technology teams to hold or grow their role in AI work, particularly around security, support, training, and integration, while 78% admit their data practices have room for improvement and 57% of those facing skills gaps plan to build training programs covering AI, data management, cybersecurity, and business process skills. Seth Robinson, CompTIA’s VP of research, summed up the finding as buying AI tools being the easy part and building an organization capable of deploying them securely and responsibly being the much bigger challenge, which is the same thing plenty of owners have discovered the expensive way. If AI is on your roadmap for the next year, the honest first step is probably a look at your data, since almost nothing useful happens on top of records nobody trusts. Read more at ISE Magazine
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 only do one thing, find out who owns the plugin and patch list for your website, because two of today’s stories start there.
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.