Back to Blog
Guides

IoT Device Security Monitoring for SMB Cameras, NVRs and Gateways

Cameras, NVRs and gateways like the UniFi consoles hit by CVE-2026-34908 sit on SMB networks with no owner and no agent. Here is how to inventory them, segment them and give a SOC the telemetry to catch one going bad.

By Danny Mercer • Nov 19, 2026 •2 views
IoT device security monitoringnetwork segmentationNVR securityManaged SOCbotnet detection
Share:

Somewhere in your building there is a camera that has never been patched, runs a password nobody remembers setting, and talks to the internet more often than your sales team does. It was installed by a low-voltage contractor during the last office remodel, it records to a network video recorder in a closet that also holds the coffee filters, and it does not appear on a single asset list your IT provider maintains. If an attacker owned it tonight, the honest answer to "how would we know" is that you probably would not.

That gap is what IoT device security monitoring is supposed to close, and for most small and midsize businesses it is wide open. Cameras, NVRs, door controllers, building management panels, smart TVs in the conference room and the consumer-grade gateway that someone plugged in to fix the guest Wi-Fi are all computers. They run Linux, they hold credentials, they reach the internet, and they almost never run an endpoint agent. They are, from an attacker's point of view, the most comfortable seats in the house.

Why IoT device security monitoring starts with the camera nobody owns

Here's the thing about unmanaged devices. The problem is rarely that a security team looked at the camera system and decided the risk was acceptable. The problem is that nobody looked at all. Physical security usually belongs to facilities, the installer belongs to a vendor, the NVR belongs to whoever has the mobile app on their phone, and the network port it plugs into belongs to IT. Four parties, zero owners.

In CISSP terms this is an asset management failure before it is anything else. NIST SP 800-53 control CM-8 asks for an accurate inventory of system components, and the moment a camera or a gateway is missing from that inventory, every downstream control quietly stops applying to it. Patch management under SI-2 cannot patch what it does not know about. Account management under AC-2 cannot rotate a default admin password it has never seen. Audit logging under AU-2 is not collecting anything from a device that was never configured to send logs anywhere. NIST wrote a whole publication on this class of hardware, SP 800-213, and its central point is refreshingly blunt. IoT devices have to be treated as system components with security requirements, not as appliances that live outside the security program.

The secondary problem is that these devices are built to be convenient. Cloud relay features, peer-to-peer remote viewing, UPnP port mapping and vendor support tunnels exist so the owner can watch the parking lot from a phone without calling anyone. Every one of those conveniences is a path in or out that the firewall rule set did not explicitly approve.

The UniFi root chain fits a decade of cameras turned into botnets

The reason this topic keeps coming back is that the attackers keep coming back. Earlier this year Ubiquiti disclosed a chain of three zero-days in UniFi OS, tracked under CVE-2026-34908, that handed attackers root on Cloud Gateways and Protect NVRs. That is a meaningful finding because UniFi gear is everywhere in the small business market. It is affordable, it looks great in a rack, and a lot of firms in the 20 to 200 employee range run their routing, their Wi-Fi and their cameras off the same console. A root compromise on the gateway is a root compromise on the network edge and the video archive at the same time.

None of this is new, which is the depressing part. Mirai tore through hundreds of thousands of cameras and DVRs in 2016 using a short list of default usernames and passwords, then pointed them at Dyn and took down a meaningful slice of the internet for an afternoon. Hikvision's CVE-2021-36260, an unauthenticated command injection in the camera web server, landed in the CISA Known Exploited Vulnerabilities catalog and was folded into Moobot within weeks. In 2021 attackers reached roughly 150,000 Verkada cameras inside hospitals, schools and jails because a super admin credential was sitting where it should not have been. In 2024 Akamai documented a Mirai variant exploiting CVE-2024-7029 in AVTECH cameras that had been end of life for years and were never going to receive a fix.

Then the nation-state crews noticed. In February 2024 the FBI disrupted a network of Ubiquiti EdgeRouters that had been compromised with default credentials by Moobot operators and then repurposed by APT28 as relay infrastructure. In September of that year the FBI, NSA and Cyber National Mission Force published an advisory on the Raptor Train botnet, run by a company tied to Flax Typhoon, which peaked at more than 260,000 devices including IP cameras, DVRs and NVRs. The pattern is consistent across all of it. Somebody else's camera becomes the attacker's proxy, and the business that owns the camera finds out from a federal advisory, if it finds out at all.

How an NVR becomes a foothold, a proxy or a weapon

Put on the CEH hat for a minute and walk the attack path. Initial access is almost always T1190, exploitation of a public-facing application, or plain valid accounts through default credentials, which MITRE files under T1078.001. The exposure usually comes from a port forward the installer created so the owner could view cameras remotely, or from UPnP doing the same thing automatically without asking anyone. Scan the IPv4 space for RTSP on port 554, the Hikvision SDK port 8000, the Dahua service on 37777 or a UniFi console on 443, and you will find small business devices by the hundred thousand.

Once the attacker has a shell, the device offers three different futures, and each one shows up differently in your telemetry.

The first future is the device as a botnet node, which MITRE describes as T1584.005. The attacker cares nothing about your business. They want your bandwidth and your residential or business IP reputation for DDoS work, credential stuffing or proxying traffic for someone else's intrusion. Your exposure here is mostly reputational and legal, plus the awkward conversation when your ISP or a federal agency calls.

The second future is the device as a pivot. An NVR on the same flat network as the accounting workstations is a Linux host with a persistent foothold that no EDR can see. From there an attacker can scan internally, relay authentication, reach file shares and wait. This is where a camera compromise turns into a ransomware incident, and the post-incident timeline usually shows weeks of quiet internal reconnaissance from an IP address that nobody recognized as a computer.

The third future is the device as a data source. Video archives contain things businesses would prefer stayed private, including badge readers, whiteboards, patient intake areas, cash handling and people's faces. Gateways like the UniFi consoles also hold Wi-Fi credentials, VPN configurations, RADIUS secrets and sometimes cloud account tokens. A root compromise there is a credential harvest, which feeds T1552 and eventually a T1078 login that looks perfectly normal everywhere else.

Asset discovery that finds what procurement never bought

Every useful IoT program begins with finding the devices, and you cannot find them by asking people. The facilities manager will remember the cameras. Nobody will remember the HVAC controller, the badge panel's embedded web server or the streaming stick behind the lobby TV.

Discovery works best from three angles at once. Start with the network itself. Your DHCP leases and your switch MAC address tables already know every device that has asked for an address, and the first half of a MAC address identifies the manufacturer. A list of MAC prefixes belonging to Hikvision, Dahua, Axis, Ubiquiti, Reolink and the usual suspects is a fast way to find hardware that never went through IT. Follow that with an active scan of internal ranges for the ports these devices like to expose.

nmap -sV -p 80,443,554,7443,8000,8443,37777 10.20.0.0/22

The third angle is the outside view, and it is the one that matters most. Search your own public IP ranges in Shodan or Censys, or have someone do it for you, and look for anything answering on video, management or remote access ports. Check whether UPnP is enabled on the edge firewall, because if it is, your inventory of port forwards is whatever the cameras decided it should be this week. CISA's BOD 23-02 was written for federal agencies, but its core idea that management interfaces do not belong on the public internet translates directly to every NVR console in a strip mall office.

Once you have the list, give every device an owner and a record in CM-8 that includes firmware version, support status and the credentials it holds. That last field is the one people skip. A camera that stores an SMTP relay password for motion alerts, or a gateway holding the RADIUS shared secret, has a much larger blast radius than its price tag suggests.

Network segmentation that assumes the camera is already compromised

Segmentation is where most of the risk reduction actually lives, and the right design principle is to assume the device is already owned. If the camera is hostile, what should it be allowed to reach? The answer is almost always the NVR and nothing else. The NVR, in turn, should reach the cameras, a time server, the vendor's update service if you rely on it and the specific workstations or app relays that view footage. It should not reach the domain controllers, the file server, the internet at large or the guest Wi-Fi.

In NIST language this is SC-7 boundary protection applied internally, and it means a dedicated VLAN for cameras, a separate one for building systems and default-deny firewall rules between them and everything else. The detail that trips people up is egress. A lot of firms put cameras on their own VLAN and then allow that VLAN full outbound internet because the mobile app stopped working when they did not. That turns segmentation into a polite suggestion. If remote viewing is a business requirement, allow it to the vendor's documented relay endpoints, or better yet put the viewing behind your existing remote access path and kill the vendor cloud relay entirely.

Disable UPnP on the edge. Remove inbound port forwards to cameras and NVRs. Change every default credential and record the new ones in the password manager under the device owner's name, which satisfies IA-5 and also means the next technician does not have to factory reset the NVR to get in. Turn off Telnet and any unused services. Set a firmware cadence for these devices just as you would for servers, and accept that some of them will be end of life with no fixes coming. The AVTECH story is the lesson there. An end of life camera does not get safer with age, so either isolate it so thoroughly that compromise does not matter or replace it.

For UniFi shops specifically, the gateway deserves to be treated as tier-zero network infrastructure rather than a gadget. Keep its management interface off the WAN, restrict local management to an admin VLAN, enforce multifactor authentication on the Ubiquiti account that controls it and verify firmware versions after patching rather than trusting the update notification. I covered the general case for keeping management planes off the internet in an earlier post on firewall management plane security, and the same logic applies with extra force when the box also stores your video archive.

What a SOC should actually watch on the device VLAN

You cannot put an EDR agent on a camera, which is exactly why network telemetry matters so much for this class of asset. The good news is that IoT devices are boring, and boring is a gift for detection engineering. A camera does a small number of predictable things. It streams video to the NVR, it checks the time, it maybe checks for firmware. Anything outside that profile is interesting.

A Managed SOC watching the device VLAN should be alerting on a short set of behaviors that are both high signal and cheap to collect from firewall logs, NetFlow and DNS. The first is any new outbound destination from a camera or NVR, especially to hosting providers, residential proxy ranges or countries where you have no business. The second is outbound connections on ports associated with scanning or spreading, like Telnet on 23 and 2323, SSH on 22 or the device-specific ports listed above, because a camera attempting to reach other cameras across the internet is a botnet node doing its job. The third is internal reconnaissance, meaning an IoT address touching SMB, LDAP, RDP or WinRM on your servers. A camera has no reason to talk to a domain controller, so a single attempt is worth a human look. The fourth is volume, since a sudden jump in outbound traffic from a device that normally sends nothing past the firewall is either a DDoS participation or data leaving.

DNS deserves its own attention. Botnet malware tends to resolve domains that look nothing like the vendor's update service, and DNS logs from the camera VLAN are tiny, so even a simple baseline of "domains this VLAN has ever queried" catches a surprising amount. If your firewall can push DNS for the device VLAN through a filtering resolver, do it, and send the query logs off the box.

That last point is the one I keep repeating across these posts because it keeps mattering. Logs that live only on the device are logs the attacker can edit. AU-9 asks for protection of audit information, and in practice that means firewall, DHCP, DNS and NetFlow records land in a SIEM or log platform that the compromised device cannot touch. Pair that with absence-of-telemetry alerting. If the NVR normally sends syslog every few minutes and goes quiet, or a camera that streamed all week suddenly stops talking to the NVR while still talking to the internet, that is a detection too.

Finally, give the SOC a playbook that is specific to unmanaged devices, because the usual endpoint playbook does not apply. You cannot isolate a camera with an EDR click. You can shut its switch port, move it to a quarantine VLAN, or block it at the firewall, and the playbook should say which of those the analyst is authorized to do at 3 AM without waking the facilities manager. It should also call for credential rotation for anything the device held, because a rooted gateway has to be assumed to have leaked every secret on it.

A practical ninety-day plan for IoT device security monitoring

None of this requires a capital project. In the first thirty days, build the inventory from DHCP, MAC tables, internal scans and an external exposure check, and assign an owner to every device. Kill UPnP, remove inbound port forwards and change default credentials. That alone removes most of the Mirai-class risk.

In the second thirty days, move devices onto dedicated VLANs with default-deny rules, restrict egress to documented destinations and get firewall, DNS, DHCP and flow logs flowing to a place the devices cannot reach. Patch what can be patched and make a replacement list for what cannot.

In the final thirty days, turn on the detections described above, baseline normal behavior per device class and test the quarantine playbook against a real camera during business hours so the first time anyone does it is not during an incident. Then put the whole thing on a quarterly review cycle, because the next contractor remodel is going to add three more devices nobody tells you about.

The UniFi zero-days, Raptor Train and every Mirai descendant all exploit the same organizational blind spot. The devices are on the network, but they are not in the security program. IoT device security monitoring is how you pull them in, and it is one of the rare areas where a small team can get most of the benefit from discipline rather than budget.

This is exactly the kind of work the Managed SOC module inside CyberSphere was built around. It watches the network, DNS and firewall telemetry that unmanaged devices cannot hide from, flags the new outbound destinations and internal scanning that mark a camera or gateway gone bad, and ties each detection to a remediation ticket in the same platform instead of a forwarded email that dies in someone's inbox. It runs 24/7/365 with dark-web breach monitoring and real-time CVE tracking on top, so the next UniFi-style advisory gets matched to the gear you actually own rather than the gear you remember buying. You can see the platform on our CyberOne page, and the partner portal can be white-labeled for MSPs who want to offer the same coverage to their own clients. The cameras are not going to start securing themselves, so somebody may as well be watching them.

Frequently Asked Questions

How quickly does Innovation Network Design respond to a security incident?

Our SOC triages and notifies within 15 minutes with confirmed details and containment steps, rather than handing you a queued alert. Incident response retainer clients carry a 2-hour guaranteed response.

Do you only work with businesses in the Dallas-Fort Worth area?

We are based in McKinney, Texas and work on-site across Plano, Allen, Frisco and the wider DFW metroplex. Our monitoring and response operate remotely, so we also support organizations nationwide.

How do we get started?

Start with a free assessment. Contact us or call 512-518-4408 and we will review your current setup and give you clear, prioritized recommendations before you commit to anything.

Happening right now?

Business email taken over, or a payment sent to the wrong account?

Call us now. The first hours matter most: your bank may still be able to recall the payment, and whoever is in the mailbox keeps access until it is locked down. We handle the lockout and help trace the payment.

Not an emergency? Get a free security assessment instead.

D

Danny Mercer

Innovation Network Design

With nearly a decade in cybersecurity and IT infrastructure, our team delivers expert insights to help businesses in McKinney, Dallas, and across DFW make informed security decisions. Have a question? Get in touch.

Ready to Secure Your Business?

Get a free security assessment and find out where your organization stands.