Author: Morey Haber, CTO, BeyondTrust
What do dolphins and digital personal assistants like Amazon Echo, Google Assistant, and Apple’s HomePod have in common? They all can hear and respond to sounds and commands outside of the range of human hearing. This can lead to some very interesting attack vectors that end users may not even know are occurring on their devices, called Dolphin Attacks. (If you’re looking for more background on Dolphin Attacks, please click here and look at what the NY Times is covering.)
If you have a personal assistant that has responded to a television advertisement or to the radio, you understand the potential risks. Dolphin Attacks take the threat to another level by issuing an inaudible command embedded in white noise or outside of the range of human hearing. While these attacks are still theoretical and in the lab for demonstration purposes, it does open a whole new world of attack vectors based on voice recognition, and we all should be concerned.
What are the risks?
First, think about a simple audible attack vector on Amazon Echo. If the device is linked to your calendar, anyone can ask for your schedule and know all your appointments. This is not a high risk at home (unless you have a snooping partner) but a high risk for an executive that may have the device in their office. If the device is at home near a window, and you have an Internet-connected door-locking system, linked to Amazon (like Nest), then a simple bang on the Window and yelling “Open the door” may allow a burglar in. And, in a recent TV commercial, a car is seen starting its engine by telling your personal assistant you are ready to leave. What could go wrong with that? Carbon Monoxide poisoning if the car is still in the garage.
The number of new threats and permutations for risk are only just beginning to be understood.
So how does this relate to Dolphin Attacks? Any one of these attack vectors could be instantiated remotely from an audible source without the end user’s knowledge. This potentially includes reconfiguring the device, making purchases, or performing other nefarious activities. If the commands are embedded in other content, the range of attack vectors as massive depending on the permissions and privileges granted to the device. This is where Privileged Access Management steps in for consumer devices.
How to solve the problem
To solve this problem, and to mitigate the potential threats from audible and inaudible commands, begin with these recommendations (provided your device supports these functions):
Enable multi user voice recognition and train the device to your voice. This will prevent unknown sources, users, and commands from executing if it does not recognize your voice.
When the device is not needed, manually mute the microphone. Many devices have a button to do so and this is a sound recommendation when you leave home, have guests over, or need privacy for conversations or events.
Do not bring these devices to your place of business. These are consumer devices and have no reason to be in your office, not even for music. If you do bring it and ignore this advice, restrict all unnecessary functions, including the calendar from being used.
Disable commands for purchasing products from the device from all but authorised users. Product purchases are an easy target for mailbox thieves.
Limit Smart Home, Mobile Phone, and Skill Access. Many of these devices can control lights, thermostats, and even make phone calls. The risks from these new features are mind boggling. If you do not need them, turn them off.
If any connections are established by your device for third parties, make sure all the passwords are unique, not re-used, and that auto updates are turned on to provide security updates automatically to prevent exploitation.
If the device supports it, add a verbal password or pin to critical commands. This will ensure purchases or configuration changes are authorized and not spoofed by a rogue command.
Basic Privileged Access Management (PAM) forms the basis for avoiding audible and Dolphin Attacks against your personal assistants. As these technologies evolve, security will get better (or worse if the glass is half empty).
Consider the permissions, privileges, and users (voices) that have access to your device. As we see these devices, and many other IoT devices appear in corporate environments, consider a commercial PAM solution to manage the threats. A PAM solution should include least privilege management through session monitoring and password management. The same risks consumers face can materialize in your corporate environment if you are embracing commercial voice recognition solutions too as a part of your IoT strategy.