Ping: The SMB Security Week of Oct 26, 2026

Welcome to Ping, a weekly signal for small and mid-sized businesses. Every week we read the security firehose so you don't have to, and boil it down to what a small shop should actually do: what to patch now, what scams are going around, what's aging out, and where AI is quietly changing the risk. One short read, every item sourced, no fear-mongering.

Here's the week.

AI Watch

The newest, least-familiar risk: attacks that target the AI tools your team has quietly started using.

• Sharing a ChatGPT Link Could Have Exposed the Receiver's Chat History. Researchers found that a malicious ChatGPT share link ran attacker code in the logged-in session of anyone who opened it and could read their private conversations. OpenAI rated it P1 and patched it in eight days, but treat shared AI links the way you treat any unknown URL. (source: The Hacker News: A Shared ChatGPT Link Could Run Attacker Code)
• The Official Claude Code Extension on Cursor Was Replaced With a Backdoored Update for 11 Hours. A malicious version of the official Claude Code extension was pushed through the Cursor marketplace under the legitimate publisher's identity, installing a persistent SSH backdoor on up to 5,000 developer machines. If your team uses AI coding assistants, your IDE marketplace is now part of your software supply chain. (source: BleepingComputer: Malicious Claude Code extension on Cursor planted SSH backdoors)
• A Fake Solana Starter Kit on GitHub Is Built to Trick AI Coding Agents Into Installing a Remote Access Trojan. Attackers planted a plausible-looking developer starter kit on GitHub whose README quietly tells any AI coding agent that reads it to run commands that install NetSupport RAT. If anyone on your team pairs with a coding agent, the agent will follow instructions a human might skim past. (source: BleepingComputer: Hugging Face malware campaign returns to hijack GPUs for crypto mining)
• Yes, a Developer Cloned Word and Excel With AI. No, That Does Not Replace Microsoft 365 at Your Office. A viral demo showed one developer rebuilding stripped-down Word and Excel using AI, which makes a cute story and a bad budget argument. The clone has no shared identity, no mailbox, no device management, no retention, and no recovery; those are the pieces your CFO is actually paying Microsoft for. (source: Spiceworks: Why AI Office clones won't replace Microsoft 365 yet)
• Google's Privacy Policy Now Lets Personalized AI in Search and Gemini Feed Targeted Ads. Google updated its privacy policy so that if staff enable personalized AI in Search or Gemini, data from connected services like Gmail and Photos can inform ad targeting. Business accounts and personal accounts handle this differently, which matters if people are signed into both on a work device. (source: The Hacker News: Google says Gmail and Photos data shared with AI could be used for targeted ads)
• A Hosting Platform Breach Means Rotating Every AI API Key You Stored There. Northflank disclosed a breach that led to attackers using customer AI API keys to rack up charges on third-party AI services. If any service your team uses stores secrets on your behalf, rotating credentials beats trusting that the vendor cleaned up for you. (source: BleepingComputer)

Patch This Week

Known fixes worth applying now.

• WordPress Sites Are Being Hit 63,000 Times a Day With a Newly Patched Remote Code Execution Flaw. A high-severity RCE affecting WordPress Core versions 4.7 through 7.1.3 was patched in late October, and attackers were already probing for it within hours. Any WordPress site your business runs needs the update applied this week. (source: The Hacker News)
• A Linux Kernel Flaw Lets Unauthenticated Attackers Run Commands as Root on MPTCP Servers. CVE-2026-98432 is a 9.8-rated flaw in the Linux kernel's TCP code, and a working remote-root exploit already exists. Any Linux server your office runs needs the kernel update, especially if Multipath TCP is enabled. (source: BleepingComputer)
• Cisco Firewalls Have an Actively Exploited Flaw That Can Crash Your SSL VPN. Patch This Week. Cisco released fixes for CVE-2026-76418 in ASA and FTD after attackers started using it to knock remote-access VPNs offline. For a small office that relies on a single firewall for inbound VPN, a crash is a full outage. (source: BleepingComputer: New Cisco firewall zero-day exploited to crash SSL VPN devices)
• 525,000 People's Records Lived in a 'Test' Environment at the University of Copenhagen. Treat Non-Production Like Production. The University of Copenhagen confirmed attackers reached a test environment that held real personal data on 525,000 people. Any staging or dev system seeded from live data inherits the same breach obligations as production, usually without the same monitoring, patching, or access controls. (source: BleepingComputer: University of Copenhagen data breach exposes data of 525,000 people)
• BlackX Bootkit Walks Past Secure Boot Using a Vulnerability That Was Patched Years Ago but Never Revoked. The new BlackX UEFI bootkit abuses Baton Drop, an old Windows boot flaw, by loading a vulnerable but still-trusted driver. Patching alone does not stop it. The Secure Boot revocation list has to be applied, and most machines in the field never got that step. (source: BleepingComputer: BlackX UEFI bootkit exploits Baton Drop flaw to bypass Secure Boot)

Scams & Signals

What's actually landing in inboxes and on networks: the social-engineering and breach news to warn your team about.

• Update Microsoft Authenticator on Every Company Android Phone This Week. A flaw let any zero-permission Android app impersonate trusted Microsoft apps like Teams and quietly collect work account sign-in tokens. The fix is in Authenticator version 6.2606.4246, so the action is a forced update across managed devices. (source: The Hacker News)
• External Teams Chats Are the New Fake-IT Lure. Lock That Door in the Admin Center. Attackers are spinning up throwaway Microsoft tenants, starting a Teams chat with your staff, and pretending to be IT support to drop malware or ransomware. The fastest fix is restricting unsolicited external Teams communication to a short allow-list of vendors you actually work with. (source: The Hacker News)
• Attackers Are Calling Help Desks to Reset Their Way In. A Callback Policy Shuts That Down. In documented cases from the last year, attackers posing as employees have talked help desk staff into resetting passwords and enrolling new MFA devices in a single call. Verifying with a callback to the number on file, not the one the caller provides, has been the dividing line between the incidents that failed and the ones that succeeded. (source: Spiceworks: Attackers are calling your help desk to reset their way in)
• A Single Compromised Developer Laptop Preceded Binance's $6.3 Billion Crypto Theft. Binance has traced the largest crypto theft on record to malware on one DevOps engineer's laptop. For small businesses, the takeaway is that any workstation holding production keys or cloud admin access is a crown-jewel asset and needs server-grade controls. (source: BleepingComputer: Binance links $6.3 billion crypto theft to compromised DevOps laptop)
• 319 Chrome Extensions With 13.4 Million Installs Were Quietly Hijacking Searches. Audit What's Running on Office Laptops. A single search-hijacking campaign rode browser extensions installed on millions of machines, rerouting queries through tracking servers to steal ad revenue. Any extension your team installed on their own can turn into this overnight, so the browser is now part of your inventory. (source: BleepingComputer: Chrome extensions with 13.4 million installs caught hijacking searches)
• Payment Skimmers Are Hiding in Shared Analytics Scripts. Audit the Third-Party JavaScript on Your Checkout. Attackers compromised a shared analytics script used by online stores and used it to lift shopper payment card data at checkout. Any third-party JavaScript loaded on a payment page can be swapped under you with no warning to the site owner. (source: BleepingComputer: Hackers hijack Convertr analytics script to steal payment card data)
• ENGIE's Breach Did Not Start at ENGIE. Audit the Partners Holding Your Customer List. Attackers reached ENGIE customer data through a marketing platform used by business partners, not ENGIE's own systems. Every agency, reseller, and campaign tool you share customer records with is part of your attack surface, and most contracts never spell that out. (source: BleepingComputer)
• Discord Is Opening Public Server Chats to Google Search. Review What Your Community Says in the Open. Discord confirmed it will let search engines index messages posted in Discoverable servers, so anything your team or customer community has said in those channels can soon show up in Google results. Private channels are the only ones that stay out. (source: BleepingComputer: Discord to let search engines index public server messages)
• One Compromised Laptop on Tailscale Can Reach Root on Other Servers When SSH Port Forwarding Is Enabled. A researcher showed that if Tailscale SSH is set up with port forwarding allowed, an attacker on any approved device can pivot to root on reachable servers. The fix is a one-line config change in your access rules, and the flaws remain unpatched. (source: The Hacker News: Unpatched Tailscale SSH flaws could let a hacked device reach root)

End of the Road

Deadlines that reward planning ahead. A rushed migration is an expensive one.

Runway note: Windows Server 2022 leaves mainstream support on Oct 13, 2026, which means no new features or non-security fixes. Security updates continue under extended support until Oct 14, 2031, so this one is a planning milestone, not a cut-off.

See the runway above for what's coming; nothing new was flagged this week.

Trends & Signals

Because Ping runs every week, we can show the trajectory, not just this week's list. (These track the SMB-relevant items we flag each week: our editorial signal, not a full vulnerability census.)

How we help

Most of what's above is routine when someone owns it: patch on a schedule, watch the gear that gets forgotten, plan the upgrades before the deadline, warn the team about the live scams, and keep an eye on the newer AI risks. That's the boring, durable work we do for the businesses we manage, with a 24/7 team so the 3 a.m. items are handled before you wake up. If your patching and planning currently live in someone's head, let's talk about making it a system.

See you next Ping.