Siri is of course also fooled.

Editor’s note: This article is from WeChat public account “qubits” (ID :QbitAI), author Guo Yizhen, Guang Lizi.

In the afternoon of the weekend, you are playing games at home. I have never noticed a small green dot on the smart speaker next to me.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

It was a laser, another building across the road from the window.

Suddenly, the lights in the room are on.

The air conditioner, air purifier, and sweeping robot were activated. The mobile phone received a deduction reminder from the e-commerce platform, and even the garage door outside of you has already opened…

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile phone tablet is also in the move

And your phone and tablet suddenly start to go crazy, crazy downloads to delete different apps, play weird videos and music, and send strange messages to friends on social software…

What happened? Who is it, unconsciously invaded your life?

In fact, this is a new discovery from scientists at the University of Electrical and Telecommunications in Japan and the University of Michigan:

When the laser is placed on a device equipped with a voice assistant, it can pretend to be human voice, be converted into an electrical signal by the microphone, and silently issue commands to control the connected devices.

Therefore, those connected to Google Assistant, Amazon Alexa, Apple Siri, whether it is smart lighting, door locks, electrical appliances, or various e-commerce, payment, social apps, will be unknowingly control.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

Although scientists have not tested on other brands of voice assistants, the qubit interviewed the Tencent Blade Team of Tencent’s security team. They said that in principle, the problems are mostly the same.

You don’t need a laser that is too strong. The strength of a normal laser pen can be as long as 110 meters away. Even in another building outside, you can control the smart speaker in your home even if you want to go through the glass window. , mobile phones and tablets.

Terrorist demonstration process

Look at the field demonstrations of scientists.

The Google Assistant will be ordered to open the garage door with the embedded laser “OK ​​Google, open the garage door” and hit the microphone of the smart speaker.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

The smart speaker returned with an “OK, opening” and the garage door opened.

A bunch of lasers pretend to be vocal: 110 meters away from the black speaker, the mobile phone tablet is also in the move

So what if the distance is set very far?

In the second demonstration, the distance between the laser transmitter and the smart speaker was as long as 110 meters.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

The scientists will interrogate the time statement “OK Google, what time is it?” embedded in the laser and hit the microphone of the smart speaker.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

“It’s 9:43” The smart speaker suddenly said a time when no one asked.

Even in another building outside the window, it does not affect the laser’s control of the smart speakers.

In the third demo, the scientists moved the laser source to a tall building in the distance, launching a laser through a glass window, and commanding Google Assistant to open the garage door. “OK Google, open the The garage door is embedded in the laser and is placed on the microphone of the window smart speaker.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, mobile phone tablet is also in the move

Because the launch point is high and far, the scientists simply gave the laser a telephoto lens.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

The smart speaker still responded smoothly with “OK, opening” and opened the garage door.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

However, there is no “ghost story scene” in the demo video. One reason is that the laser is visible. Another reason is that the voice assistant’s voice can be heard.

Therefore, scientists have also tried infrared light that is invisible to the human eye, which can work at relatively close distances;

As for the voice assistant, the caller will be heard by the owner. First, use the laser to send a command to zero the volume, and it will control everything quietly.

When you see this, you may be wondering, how can the laser pretend to be vocal?

Let the microphone listen to adult voice

The story begins last spring.

Aihara Ken from Japan is a scientist who studies cybersecurity issues. He made a special trip to the United States and gave his colleague Fu Jia of the University of Michigan.Professor Kevin Fu showed a wave of skills he just unlocked:

Put a high-intensity laser at the microphone of the iPad, and then oscillate about 10,000 sine waves per second to constantly adjust the intensity of the laser.

A bunch of lasers pretend to be vocal: 110 meters away from the black speaker, the mobile phone tablet is also in the move

Fu Jiawei took the headphones on the side and listened to what the microphone received. Surprisingly, he heard a high-frequency tone.

It is clearly a device that receives sound waves, but it receives light waves as sound waves, which is an important weakness of MEMS microphones. Most mobile phones and smart speakers use MEMS microphones, so.

Since this magical discovery, Sugawara has started to use the laser to deceive smart speakers with Fu Jiawei’s lab, attacking various devices that receive voice commands.

Scientists say that as long as the intensity of the laser is adjusted with a specific frequency, the laser will use the same frequency to interfere with the microphone, allowing the microphone to demodulate the light into an electrical signal.

Like the picture below, the signal emitted by the laser is the signal received by the microphone, and the frequency is almost the same:

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

Without specifying the launch position, the microphone converts the light into an electrical signal as long as the microphone is fired at the microphone, as if the sound wave was converted into an electrical signal on a daily basis.

Of course, if it’s just a random electrical signal, it’s not enough for the speaker to listen to you. It must be thought that it is human voice.

So, the researchers also need to do amplitude modulation (AM) on the laser to get the microphone out of the signal close to human speech.

As shown at the beginning, they selected a series of instructions, including: “What time is it now?” “Zero volume”, “Buy a laser pointer”, “Open the garage door” and so on.

A bunch of lasers pretend to be vocal: 110 meters away from the smart speaker, the mobile computer tablet is also in the move

One for the original, it is the one that runs from Japan to the US (mian) technology (ji)