Good morning. Yesterday was Patch Tuesday, and it was the biggest one Microsoft has ever shipped, which sounds alarming until you read the part where the researchers themselves say the pile is getting taller without getting much more dangerous. That tension runs through most of today’s news, because AI is now producing both the volume of findings your team has to sort through and the speed at which attackers move through what nobody sorted. Here are the five stories that matter most if you are running a small or midsize business, or leading its IT.
Microsoft shipped the largest Patch Tuesday in its history, and the useful takeaway is not the number. The September release fixed 974 CVEs, including two privilege escalation flaws already being exploited in the wild, CVE-2026-85880 in the Windows Advanced Local Procedure Call component and CVE-2026-81963 in the Windows Update Stack, both of which let a local attacker reach SYSTEM. Twenty more are considered wormable, meaning they allow remote code execution with no login and no user interaction, spanning DNS Server, DHCP, Netlogon, Active Directory and SMB Client, and the fixes also cover Exchange, SharePoint, SQL Server and Remote Desktop Services. Tenable’s Satnam Narang put the important framing on it when he said AI-assisted vulnerability discovery is creating larger haystacks without finding more needles, and that the number of flaws that actually affect a given organization remains quite low. That is the sentence to carry into your next conversation about patching, because a 974 headline invites either panic or paralysis, and neither is the right answer. What you want from your team or your provider is not a promise to patch everything this week but a defensible answer to which of these are reachable in your environment and which got fixed first. Read more at SecurityWeek
SAP disclosed a maximum-severity flaw in the code underneath nearly everything it sells, and more than 10,000 internet-facing systems are potentially exposed. CVE-2026-44756, named OVERPASS by the Onapsis researchers who found it, is a buffer overflow in the Extended Passport Protocol library inside the SAP Kernel, and successful exploitation lets an unprivileged attacker run commands with administrative rights on the SAP host, which means full compromise of the business data sitting behind it. It is reachable through the Internet Communication Manager, the component that connects an SAP application server to the outside world, and Onapsis calls its 10,000 system estimate conservative because the SAP Web Dispatcher is hard for internet scanners to fingerprint. SAP fixed 20 issues in total on Tuesday, including a second critical one called S4GET in the NetWeaver Message Server that cannot be firewalled off without breaking SAP GUI logins for your users. If you run any flavor of SAP, this is a stop-what-you-are-doing item rather than a next-cycle item, and the wider lesson holds for the rest of us, because the systems most likely to hurt a midsize company are the ones that quietly do everything and are therefore the hardest to take offline for an hour. Read more at BleepingComputer
Google’s threat intelligence team documented attackers handing entire attack campaigns to AI agents, including one that stole thousands of credentials in under six hours. The Q3 2026 AI Threat Tracker from Google Threat Intelligence Group draws on Mandiant incident response work and describes a shift away from asking a chatbot for help toward multi-agent frameworks that plan, execute, troubleshoot and adapt on their own. In one Q2 investigation, a financially motivated actor compromised a company’s cloud infrastructure, deployed an autonomous framework built from a coding chatbot and a set of agent instructions, and ran a mass credential harvesting campaign end to end in less than six hours, with the AI managing the scanning pipeline, fixing its own errors and rotating IP addresses while routing traffic through the victim’s own legitimate cloud addresses. Google also found an exposed attacker server running a dashboard that validated and organized more than 23,800 stolen secrets in real time. The company says it has not yet seen a fully autonomous pipeline used against a target, and that caveat matters, but the direction is clear enough to plan around. The practical read for a smaller company is that the window between a credential leaking and that credential being used is collapsing, so the controls worth funding are the ones that assume compromise and shorten your reaction time, like short-lived credentials, alerting on unusual sign-ins, and knowing who can revoke a token at nine on a Friday night. Read more at Help Net Security
French AI lab Mistral raised three billion euros at a valuation above 21 billion, and the reason it could is the part worth your attention. The Series D, announced Tuesday and led by Samsung Electronics with EQT’s Scaleup Europe Fund and PSG Equity as co-leads, is the largest equity round ever completed by a European technology company, and the money goes toward compute capacity, infrastructure and international expansion across the 20 countries Mistral now operates in. What sets the company apart is not model benchmarks but positioning, since it sells to governments and corporations that want control over where their AI queries are processed, it began letting customers pick the region their inference runs in, and it hosts third-party open-weight models so buyers are not locked to one provider. Investors are paying a premium for that, which tells you data residency and provider independence have moved from a compliance checkbox to something with real money behind it. If your business handles regulated data, has customers in Europe, or has simply started wondering what happens when your AI vendor changes its terms, it is reasonable to start asking where your prompts are processed, who retains them, and how hard it would be to move to a different model. Those questions are becoming standard, and vendors are increasingly expected to answer them. Read more at TechCrunch
A security firm just gave away an agentic AI platform for security operations work, and free tooling in that category is a signal about where your security spend is heading. Security Risk Advisors launched SCALR AI on Tuesday, delivered through Amazon Marketplace with no license fee and deployable inside a customer’s own Azure tenant, letting security teams build multi-agent workflows for incident triage, threat hunting, vulnerability management and phishing analysis while paying only for the compute they use. The platform runs specialized agents that react to events rather than waiting for someone to type a prompt, includes an AI quality review step on the agents’ own output, and shows cost per workflow, per agent and per model, which is a fairly honest acknowledgment that AI security tooling has a runaway spend problem. Whether or not you ever touch this product, the thing to notice is that the economics of monitoring and triage are changing fast, and if your security is outsourced you should expect that to show up in what you get rather than only in someone’s margin. The fair question to put to a provider is not whether they use AI, since everyone now says yes, but which decisions the agents are making, what a human reviews before anything acts on your network, and what you are being charged for. Read more at ChannelE2E
Follow Up
Follow-up to yesterday’s N-able story: CISA has now added that flaw to its Known Exploited Vulnerabilities catalog with a September 11 deadline. CVE-2026-86218, the CVSS 10.0 static code injection flaw in N-able N-central that allows pre-authentication remote code execution, was one of four vulnerabilities added Tuesday on evidence of active exploitation, alongside the Adobe Commerce and Magento zero-day and both of the Windows zero-days from this month’s Patch Tuesday. Federal civilian agencies have until Friday to patch, which is the shortest kind of deadline CISA issues and settles the question of whether the attacks N-able described privately to customers were real. If you outsource your IT and have not yet heard from your provider about N-central, today is the day to ask, since a KEV listing with a three-day federal clock is not a maintenance-window item. Read more at CISA
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 with this edition, ask whoever handles your patching which of these fixes actually apply to 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.