Live data from Hacker News

Using ultrasound attack to disarm a smart-home system

theregister.com

21–30 of 37 posts

Re: Using ultrasound attack to disarm a smart-home system

#21
post #11

I simply do not expose critical switches to Siri from Home Assistant.

I only have Siri kick in when I press the button.

I disabled it after a very potentially awkward event. I was talking to my wife while Siri on my Apple Watch misunderstood me, triggered itself and sent a text saying "I love you" to our previous maid (which we had fired because of reasons).

Re: Using ultrasound attack to disarm a smart-home system

#23
post #7

> could be remotely hijacked, using carefully crafted near-ultrasonic sounds, and forced to make unwanted phone calls and money transfers, disable alarm systems, or unlock doors. I know Siri is much too dumb to facilitate a money transfer…

At least with Alexa devices disabling alarm systems and unlocking doors requires a PIN, mostly to defend against the much more low tech attack of shouting through the victim's letter box.

Re: Using ultrasound attack to disarm a smart-home system

#25
post #2

The article taught me that I can control Siri’s feedback volume by for example saying “Hey Siri, speak 25 percent.”

I learned that too, but now I'm wondering why her voice suddenly went silent a few weeks ago...

This has been driving me mad. It seems certain actions have had all audible feedback disabled, but there's no real sense to which ones.

When I say across a room to set a reminder or add something to my shopping list my Homepods will just silently do so, with no indication but a flash of the screen. I have no idea if it's registered what I was saying or not.

When I ask to turn on the lights in a room, it'll do a bing-bong noise at me to indicate that it's registered despite the fact I can see the lights turning on. It's utter nonsense.

Re: Using ultrasound attack to disarm a smart-home system

#26
post #5

Why the heck wouldn't they have the wake-word listener confine itself to human-audible frequency ranges? Seems like that would be a really simple fix with zero loss of real-world functionality....

Yeah, it's weird. There was already a scare about ultrasonics being used for ad tracking in 2016 [1] so it's not like this is an unknown attack vector, and I thought the subsequent patch efforts had already added filters that stopped the phones listening on ultrasonics.

I also remember seeing a presentation on the first gen Echo which went into its noise cancelling tech, making sure that stuff coming out of the speaker wasn't received by the mic, so the success of the speaker-to-mic attack vector also seems totally bizarre.

1: https://www.wired.com/2016/11/block-ultrasonic-signals-didnt...

Re: Using ultrasound attack to disarm a smart-home system

#27
post #8

> It's also worth noting that the length of malicious commands must be below 77 milliseconds — that's the average reaction time for the four voice assistants across multiple devices. I don't get it. Why? You can speak to smart assistants much longer than that isn't it?

This har me confused at first too. Basically you are playing audio on the device and that audio will send a command to the same device. If I’m watching a video on my phone and say hey siri, it will interrupt the audio playing from my phone as it listens to me. But it isn’t instantly interrupting the audio, it apparently takes about 77 milliseconds.

Ah I see, thank you!

Re: Using ultrasound attack to disarm a smart-home system

#28

Earlier quoted context omitted.

I only have Siri kick in when I press the button.

I disabled it after a very potentially awkward event. I was talking to my wife while Siri on my Apple Watch misunderstood me, triggered itself and sent a text saying "I love you" to our previous maid (which we had fired because of reasons).

Years ago a classmate in college was showing Siri to one of our teachers. As a joke he said "Siri, I love you!" and Siri replied "I'm sorry, I can't find a location for ".

Re: Using ultrasound attack to disarm a smart-home system

#30
post #17

> It's also worth noting that the length of malicious commands must be below 77 milliseconds — that's the average reaction time for the four voice assistants across multiple devices. I don't get it. Why? You can speak to smart assistants much longer than that isn't it?

What I'm wondering is how a useful command can be expressed in 77 ms?

"weather"
Post reply on HN