Updated for the missed 5 PM Eastern recap on July 27, 2026. Yesterday’s scheduled recap did not appear on WordPress, so this repost catches up the practical stories that mattered most Monday evening: an actively exploited SD-WAN management flaw, a Fortinet FortiOS issue added to CISA’s exploited-vulnerability catalog, phishing that abuses fake Teams updates to install remote-control tools, a large router and IoT botnet, and the industry push to make AI agents easier to secure.
Quick Take
- Bad news: CISA added two exploited vulnerabilities on July 27: Fortinet FortiOS CVE-2025-68686 and Arista VeloCloud Orchestrator CVE-2026-16812.
- Bad news: Arista’s on-premises VeloCloud Orchestrator flaw is a maximum-severity unauthenticated command-injection issue with active exploitation reported.
- Bad news: Operation BlueDash used fake Microsoft Teams update pages to install legitimate remote monitoring tools for attacker access.
- Mixed news: AI agent security is moving from warning to engineering response, with NVIDIA and partners forming an Open Secure AI Alliance.
- Good news: The most useful fixes are still practical: patch exposed management tools, restrict admin portals, remove default passwords, and control who can install remote-access software.
1. CISA Added Fortinet And Arista Flaws To The Exploited List
CISA’s Known Exploited Vulnerabilities catalog added two entries dated July 27, 2026. The first is CVE-2025-68686, a Fortinet FortiOS information-exposure issue. CISA’s description says it can let a remote unauthenticated attacker bypass a patch related to a symbolic-link persistence mechanism observed in some post-exploit cases, although the attacker would first need filesystem-level compromise through another vulnerability.
The second is CVE-2026-16812 in Arista VeloCloud Orchestrator on-premises. This one deserves immediate attention for any business or service provider using affected SD-WAN management infrastructure. CISA gave federal agencies a July 30, 2026 due date, which is unusually short and a useful signal for everyone else: exposed network-management systems should not wait for the next normal maintenance window.
For home users, these specific enterprise products may not be in the house. For small businesses, MSPs, and offices with managed network gear, the lesson is broader: internet-facing management portals, firewall appliances, VPNs, SD-WAN controllers, and router admin tools are high-value targets. If one of those systems is compromised, the attacker may be able to see configuration, credentials, certificates, managed devices, and internal network details.
- Inventory Fortinet, Arista, VeloCloud, firewall, VPN, and SD-WAN products before assuming this does not apply.
- Patch affected systems and check vendor advisories for fixed versions and compromise guidance.
- Restrict admin interfaces to VPN or administrative networks wherever possible.
- Review recent administrator activity, device configuration changes, certificate access, and unusual outbound traffic.
- If exploitation is suspected, preserve logs before rebuilding or wiping the system.
2. Arista’s VeloCloud Orchestrator Bug Is A Serious SD-WAN Warning
BleepingComputer reported Monday evening that Arista patched an actively exploited, maximum-severity command-injection vulnerability in on-premises VeloCloud Orchestrator deployments. The affected platform is used to configure and manage VeloCloud SD-WAN deployments and edge devices, so compromise is not just a single-server problem.
Arista’s advisory says the vulnerability can let remote attackers reach privileged functionality intended only for internal use. BleepingComputer reported that affected on-premises versions include several 5.2.x, 6.1.x, 6.4.x, and 7.0.x builds before the fixed releases. Hosted and dedicated deployments were patched before the public advisory, and VeloCloud Gateway and Edge products were not listed as vulnerable to this flaw.
That matters for offices because SD-WAN is often treated as boring plumbing once it works. It should not be invisible from a security standpoint. The controller may know about sites, tunnels, edge devices, certificates, credentials, and routing. If an attacker has already touched that system, applying a patch may not be enough by itself.
- Ask your MSP or network vendor whether you use VeloCloud Orchestrator on-premises.
- Confirm the exact version and whether it is one of the affected release trains.
- Check whether the management interface is reachable from the internet.
- Rotate credentials and certificates if compromise is suspected.
- Review edge-device configuration after the controller is secured.
3. Fake Teams Updates Are Being Used To Install Remote-Control Tools
The Hacker News covered a campaign called Operation BlueDash that used Microsoft Teams-themed lures and fake “secure document” flows. Victims were sent through compromised infrastructure to a counterfeit Microsoft Store-style page that claimed Teams needed an update before the shared document could be opened.
The payload installed legitimate remote monitoring and management tools, including Level RMM and ConnectWise ScreenConnect. That is exactly why this kind of attack is dangerous: the software may be real, signed, and useful in normal IT support, but when it appears from a phishing page it becomes attacker-controlled remote access.
For a small business, this should tighten the rules around support tools. Staff should not install a remote-support agent because a document page, meeting page, or browser pop-up says it is required. Real support tools should come from your IT provider through a known process, and installed agents should be visible in an inventory.
- Train staff to update Teams from Microsoft Store, Windows Update, or the official Microsoft Teams app, not from document links.
- Keep an approved list of remote-access tools used by your business.
- Alert on new ScreenConnect, AnyDesk, TeamViewer, Level RMM, Tactical RMM, MeshAgent, and similar installations.
- Remove stale remote-support agents left behind by old vendors.
- Require a known callback path before allowing a remote technician into a business machine.
4. The Dysphoria Botnet Is Another Router And IoT Hygiene Reminder
BleepingComputer reported that a botnet called Dysphoria has compromised roughly 200,000 devices worldwide and is being used for DDoS attacks and traffic relay operations. Researchers tied the spread to weak Telnet and SSH credentials plus known vulnerabilities in routers, cameras, and other IoT devices.
The technical details are worth noticing. The botnet uses blockchain-based infrastructure discovery, including Ethereum ENS and Solana SNS domains, which can make command-and-control infrastructure harder to track and disrupt. XLab also observed a variant focused on turning infected devices into network proxies instead of only DDoS nodes.
This is not just an enterprise problem. Home routers, small-office cameras, old Wi-Fi extenders, DVRs, and exposed admin panels are common botnet targets. If a device still has a default password, old firmware, UPnP enabled unnecessarily, or internet-facing Telnet/SSH, it can become someone else’s attack infrastructure.
- Change default admin passwords on routers, cameras, NAS devices, and Wi-Fi gear.
- Disable remote admin access unless there is a documented need.
- Turn off UPnP when it is not needed.
- Update firmware on routers and cameras, or replace devices that no longer get updates.
- Put cameras and IoT devices on a guest or isolated network where practical.
5. AI Agent Security Is Getting More Concrete
The Hacker News reported that NVIDIA and 36 other organizations formed the Open Secure AI Alliance, with members across cloud, security, enterprise software, AI companies, and open-source organizations. The stated scope includes identity, permissions, isolation, guardrails, logs, model formats, multi-model scanning, and secure coding workflows.
That is the right set of words because agent security is no longer just about whether a chatbot says something wrong. AI agents can read files, call tools, browse repositories, run code, and act across connected systems. The practical security question is: what can the agent touch, what instructions can influence it, what logs exist, and what happens if it is tricked?
NVIDIA also released the NOOA research framework under Apache 2.0, but its own warnings matter: tools that execute model-generated Python need real containment. Tracing and inspection help, but a container, virtual machine, or operating-system sandbox is the real boundary when generated code can run.
- Do not give AI agents broad access to production secrets.
- Use separate test accounts and limited-scope tokens for agent experiments.
- Review repository instruction files and pull-request comments as possible attack inputs.
- Log agent actions, tool calls, file access, and external requests.
- Keep human approval for actions that send messages, spend money, change accounts, publish content, or touch customer data.
6. Google AI Overviews Are Changing How People Find Answers
TechCrunch reported Monday that Google’s AI Overviews are appearing in a much larger share of searches, citing data that moved from 15% to 43% over a year. For normal users, this changes how answers are consumed. For small businesses, it changes how content needs to be written.
The useful response is not to chase gimmicks. Businesses should publish clear, specific, local, source-backed pages that answer real customer questions better than a generic AI summary can. That means practical details, service-area context, pricing or process expectations when appropriate, original photos where useful, and clear next steps.
- Make service pages answer actual customer questions, not just list keywords.
- Add practical troubleshooting and buying guidance where it helps readers.
- Use clear titles, headings, internal links, and source links.
- Keep company contact and appointment paths easy to find.
- Monitor search snippets and indexed pages for spam or unexpected titles.
Today’s Practical Checklist
- Patch exposed infrastructure: especially Fortinet, Arista VeloCloud Orchestrator, VPN, firewall, SD-WAN, and remote-management systems.
- Control remote-support tools: know which RMM agents belong on each machine and remove anything unauthorized.
- Clean up routers and IoT: update firmware, disable unnecessary remote access, and replace unsupported devices.
- Secure AI agents: limit permissions, isolate execution, log actions, and keep approval gates for sensitive work.
- Write for real customers: as AI search grows, detailed and trustworthy local content matters more, not less.
If you want help checking your firewall, router, website, Microsoft 365 security, remote-support tools, backups, or AI-tool permissions, schedule an appointment with The IT Guys. We help home users and small businesses turn these headlines into practical fixes.
Sources
- CISA Known Exploited Vulnerabilities Catalog
- Arista Security Advisory 0144 for VeloCloud Orchestrator
- Fortinet PSIRT FG-IR-25-934
- BleepingComputer: Arista patches VeloCloud Orchestrator zero-day
- The Hacker News: Operation BlueDash fake Teams update campaign
- BleepingComputer: Dysphoria DDoS botnet
- The Hacker News: Open Secure AI Alliance and NOOA framework
- TechCrunch: Google’s AI search is rapidly becoming the default
- The Hacker News: vBulletin pre-authentication code execution exploit details