November 19, 2026
Learning IoT Security with the Nexsmart Guardian 2
A wireless alarm system, some inexpensive radio hardware, and a surprisingly long introduction to coordinated vulnerability disclosure.

By Elias Lindström
8 min read
My investigation of the Nexsmart Guardian 2 uncovered three vulnerabilities: replaying wireless control commands, inferring household activity from sensor transmissions, and suppressing notifications during a temporary network interruption. Together, these allowed me to disarm the tested system without the usual notification reaching the user.
Nexsmart has since taken steps to phase out the affected models, disclose their RF limitations to prospective buyers in Denmark, and offer refunds to concerned customers. The vulnerabilities have been assigned CVE identifiers, with publication of the records being coordinated.
That is the short version. The longer version involves considerably more waiting for electronics to arrive.
Introduction
A while back, a friend introduced me to the Nexsmart Guardian 2. It is a modular alarm system that allows users to pair door sensors, motion sensors, keypads, and other accessories with a central "brain." My friend was happy with the installation process and the price, but at what cost was the convenience?
I have a background in cybersecurity competitions, but investigating a real product was something different: No intended solution and no reassuring flag to search for. I searched for existing security research on the device but found little to work from. An interesting project to be sure. All hands-on testing described here was local, involving the alarm hardware, its RF communications, and the local network. I did not conduct security testing against Nexsmart's or Tuya's servers.
Initial investigation
I could not find a copy of the firmware online, and I did not want to start by taking the device apart. That left me with a black box and whatever I could learn from its external interfaces. Luckily, the product information revealed two obvious attack surfaces to examine: a Wi-Fi connection and a 433 MHz radio link to the sensors.
For the radio side, I needed something capable of receiving and transmitting on the relevant frequency. I found a CC1101 radio transceiver that I could connect to an ESP32 I already owned. This seemed like an economical way to get started, provided I valued my troubleshooting time at approximately zero.
I also picked up a Wi-Fi adapter I had been eying for a while, capable of monitor mode and packet injection. Research and the common good of securing consumers' homes turns out to be an excellent justification for buying what you want.
While waiting for the hardware, I investigated how the alarm communicated with its app. The Guardian 2 uses Smart Life, part of the Tuya IoT platform. That gave me a useful starting point: although I had found little research specifically about this alarm, there was considerably more information available about Tuya.
Investigating the Wi-Fi communication
My first approach was to inspect traffic between the alarm, the Smart Life app, and the cloud. Simply observing the connection through ARP poisoning did not reveal much. The cloud traffic I inspected was protected by TLS, including MQTT over TLS. MQTT is a messaging protocol commonly used in IoT systems; TLS provided the encryption that prevented me from reading the interesting parts.
The encryption was doing its job. Inconvenient for me, reassuring for everyone else. The traffic did reveal something useful, though. When the phone running Smart Life and the Guardian 2 were on the same local network, they could communicate directly instead of sending every command through the cloud.
Other researchers had already investigated Tuya's local protocols, and projects such as TinyTuya provided a much better starting point than beginning from scratch. My inspection identified the device's local communication protocol as Tuya version 3.4. This is where I learned an important lesson about checking what research already exists.
Naturally, I also spent time setting up an Android emulator, installing and decompiling Smart Life, and experimenting with static and dynamic analysis. Inspecting the app's encrypted traffic required additional configuration, and the application protocol introduced further encryption and key-handling details.
It was an interesting detour into mobile reverse engineering. It was also a very effective way to spend several evenings without getting much closer to my original goal. Eventually, I returned to the question I had actually set out to answer: how secure was this particular alarm system?
The RF experience
I was pretty new to RF work, so getting the CC1101, ESP32, and computer to cooperate took some time. A small radio transceiver is not quite the same thing as having a full view of the surrounding radio spectrum. I had limited insight into what was reaching the antenna and plenty of opportunities to wonder whether the problem was my code, wiring, configuration, or understanding of physics.
After spending enough time with the microcontroller and the datasheet, I started decoding signals from the sensors. Comparing their timing and structure with documented protocols gave me a likely protocol family and an educated guess about the encoder chip involved.
The more important discovery required much less speculation: the messages were static. Opening a door caused its sensor to transmit a particular code. Closing it produced another. Repeating the same action produced the same transmission. This meant that someone listening nearby could associate recurring messages with sensor events. Over time, those events could reveal patterns consistent with people arriving, leaving, or moving around the property. This finding was assigned CVE-2026–82949.
The listening itself was entirely passive. There was no way for the victim to know they were being observed through these transmissions. It was already an interesting result, but the keypad raised a more immediate concern. Its transmissions followed a similar pattern. Instead of reporting a door event, they requested actions such as arming or disarming the alarm. These commands were static too.
After observing legitimate keypad traffic, I tested whether the system would accept a replayed control message. It did. I could arm and disarm the tested system without entering the keypad code. The alarm accepted the wireless command without an effective check that it was fresh and came from the legitimate controller. This finding was assigned CVE-2026–82948.
I was excited about the discovery. My friend was understandably less enthusiastic about this particular demonstration of value for money. He also pointed out an important limitation: changing the alarm's state still generated a notification in his app. Even if someone could disarm it, he would know that something had happened. That was a useful protection. The next question was whether it still worked when the alarm temporarily lost its network connection…
When notifications disappear
The earlier Wi-Fi investigation now became relevant. The Guardian 2 needed its network connection to deliver updates locally or through the Tuya cloud. Interrupting that connection could therefore reasonably be thought to interrupt notification delivery.
One way to cause such an interruption is a Wi-Fi deauthentication attack. This abuses management frames used to end a wireless connection. On connections without effective Protected Management Frames (which are a large portion of the personal network), forged frames can cause a device to disconnect even when the sender does not know the Wi-Fi password. According to a 2025 report from Nozomi Networks roughly 94% of home networks do not have PMF enabled.
I also investigated whether observable traffic patterns could help distinguish the alarm from other wireless devices. The device's periodic Tuya discovery traffic provided useful clues, even without reading its encrypted contents. Those clues were not a unique identity card for the Guardian 2. Other Tuya devices could produce similar traffic, so identifying a particular alarm would require context and could involve uncertainty.
The significant result was what happened to notifications during the interruption. I expected events to be retained and delivered after the alarm reconnected. In my tests, the relevant notifications did not arrive after connectivity returned.
Because this was a black-box investigation, I could observe the delivery failure without conclusively identifying which component failed to retain or deliver the event. The practical outcome was clear: a security-relevant state change could happen during a temporary disconnection without its notification subsequently reaching the user.
The system did provide an offline alert after approximately 30 minutes of disconnection in my testing. That is useful, but it addresses a different problem from losing individual notifications during a shorter interruption. This finding was assigned CVE-2026–82950.
Deauthentication is a well-known technique rather than a discovery specific to this alarm. The product-specific concern was the loss of security notifications across a temporary outage. Connection protection matters, and an alarm also needs to handle interruptions safely when they happen.
Combined with replaying a control command, the notification failure removed the warning my friend had initially relied on. For this public writeup, I am keeping the practical targeting details and the precise sequencing of the combined attack out of scope. The important result is that the vulnerabilities could reinforce one another.
The exciting bureaucracy
With the research complete, it was time to tell the vendor. As these were my first findings to go through coordinated vulnerability disclosure, I began by writing a technical report explaining the vulnerabilities, their impact, and possible mitigations. I contacted Nexsmart and followed up but initially received no substantive response.
I eventually asked the Swedish national CSIRT, CERT-SE, to help establish contact, and hats off to them. They were excellent to work with. CERT-SE tried several routes and ultimately received assistance from Forsvarets Efterretningstjeneste, the Danish Defence Intelligence Service, in reaching the company.
I had expected to learn something about radio protocols. The organizational chart of Danish cybersecurity was an unexpected bonus. Once contact was established, Nexsmart explained that the earlier messages had been classified as spam and closed without being escalated internally. They acknowledged the reports, began an investigation with their supplier and manufacturer, and became constructive to communicate with.
I agreed to extend the disclosure timeline while that work continued. Nexsmart initially reported suspending sales. In a later update, the company described a phase-out of the current Guardian models: remaining stock would be sold only in Denmark, the products would not be reactivated in other markets, and no further orders would be placed for those models.
The Danish product pages were also updated to explain that the RF system uses fixed, unencrypted codes without rolling-code protection. That limitation is now acknowledged on the Guardian 2 product page. For existing customers, Nexsmart said it would offer refunds to those who contacted the company with concerns about the identified security limitations and no longer wished to keep their system. The offer included purchases made more than 100 days earlier. They are also developing a newer version to replace the Guardian 2 system.
The company has expressed support for public disclosure and asked to review the writeup for factual input and to ensure its response is represented fairly. Publication of the CVE records has been coordinated through CERT-SE and the CNA is ENISA.
What this means for owners
The practical concerns are nearby attackers replaying RF control messages, observing sensor activity, and interrupting delivery of security notifications. The demonstrated disarming scenario involved a paired RF controller, such as the keypad tested here, and observation of legitimate controller traffic.
A sales phase-out, a product-page warning, and a refund offer do not themselves correct the vulnerabilities in an installed system. If you own an affected system and are concerned, contact Nexsmart through its official support channels and ask about the refund offer and any current remediation guidance. The offer described above comes from the company's disclosure correspondence, so confirm the applicable terms directly.
I would also avoid assuming that changing how you arm the system, or changing a router setting, resolves all three findings. These are separate problems, and a change that helps with one may leave the others untouched.
What I learned
The most useful lesson from this project was how much of security research happens outside the moment when an exploit works.
It involved reading existing research, choosing an attack surface, questioning assumptions, learning unfamiliar tools, and separating an observation from an explanation of what caused it. It also involved writing reports and helping people in different organizations understand the same problem. And waiting for electronics to arrive. Quite a lot of that.
The initial Wi-Fi investigation felt unproductive until it helped explain a later finding. The cheap radio hardware was enough to uncover meaningful problems, although the money saved came with an impressive troubleshooting subscription.
Most importantly, looking at each protection separately would have missed the full picture. A replayed command still produced a notification. A temporary network interruption appeared temporary. Together, they exposed a more serious weakness.
That made this an interesting first IoT research project, and a much broader introduction to disclosure than I had expected when I first looked at a small alarm plugged into a wall.