The proliferation of Internet of Things devices seems to be unstoppable. However, along with the increase in the number of devices out there comes the security risk that these sensors and connected things can pose when they are compromised by an attacker.
In addition, many security tools don't adapt well to IoT. (See Increased IoT Use Causing Added Enterprise Security Concerns – Report.)
These tools have historically been based on assumptions about the protocols that are used in the network connections and the device states that exist when they attach to the network. For instance, IoT devices are too variable in these areas to fit into neat categories.
Now, however, researchers have come up with a self-learning system aimed at detecting compromised IoT devices, which does not require any prior knowledge about the device types or require pre-programmed training data to operate.
Here's how they describe it:
We propose a novel approach that combines automated device-type identification and subsequent device-type-specific anomaly detection by making use of machine learning techniques. Using this approach, we demonstrate that we can effectively and quickly detect compromised IoT devices with little false alarms. [It] is completely autonomous and can be trained in a distributed crowdsourced manner without requiring human intervention or labeled training data.
Sounds too good to be true, but the researchers say that they have it.
Giving it the name of DÏOT, the system has two main components.
The first is the "Security Gateway" and the second is the "IoT Security Service." Together, these two components detect compromised IoT devices by monitoring their communication as observed by the Security Gateway, which acts as a network gateway for the local network.
The security service also has cloud-based functionality, which has two main components: Device-Type Identification and Anomaly Detection Model.
The security service trains the gateway by using fingerprints that are generated at several Security Gateways to learn the specific device-type identification models that are attached to the network. The aggregating maximizes the usage of limited information obtained from scarce communications at each gateway.
The Anomaly Detection Model maintains a repository of device-type-specific anomaly detection models which are matched to the signatures gathered.
Once a model is chosen, the system looks at the current traffic pattern to see if it matches the normal pattern expected. This is done through the use of neural network techniques. If it doesn’t matchup, an anomaly alert is generated.
The fundamentals of network security are being redefined -- don't get left in the dark by a DDoS attack! Join us in Austin from May 14-16 at the fifth-annual Big Communications Event. There's still time to register and communications service providers get in free!
The system was tested on a lab network that had several devices, such as appliances, smart lamps, cameras, and routers connected to it by WiFi. The network was allowed to connect for 24 hours before it was tested. It turned out that effective device identification for 33 different IoT devices required only a few hours of traffic monitoring.
Various attacks were then implemented, most based on the Mirai botnet. (See IoT Malware-on-the-Fly Expected to Rise .)
Researchers found that most aggressive distributed denial of service (DDoS) attacks were detected in one millisecond with 100% accuracy. Overall, it detected 96% of attacks in less than 0.03 seconds with a low false alarm rate of 1%.
This kind of system has some potential huge benefits.
Mostly, it is automated and effective and so can serve as a protective barrier that can be widely implemented. Further development with this technique may help to rid the Internet of the dangers of malicious IoT devices and the bots that go with them.
— Larry Loeb has written for many of the last century's major "dead tree" computer magazines, having been, among other things, a consulting editor for BYTE magazine and senior editor for the launch of WebWeek.