The heart of this particular thing/story was a spoken phrase, a phrase which you should avoid speaking out loud if you are within hearing distance of an Amazon Echo device like the one on the right. The phrase is: "Alexa, order me a dollhouse."When the morning TV news program on station CW6 in San Diego reported that a young girl had accidentally used her parents' Amazon account to purchase a very expensive dollhouse via Alexa, the news anchor Jim Patton said: "I love the little girl saying ‘Alexa order me a dollhouse.’” As soon as Jim said that, the phones at the TV station started ringing. Viewers were calling to complain that their Alexas had tried to order dollhouses. In other words, a whole lot of people had been awoken to the fact that the current generation of Alexa devices will take orders from anyone: they use voice recognition technology to understand what people say, but not to distinguish who is saying it.
Later that Thursday morning, CW6 called ESET, the US headquarters of which are in San Diego, and asked if I could comment on this phenomenon. I said yes because I was already doing research on digital devices with voice recognition including, oddly enough a doll called "My Friend Cayla". Reporter Carlos Correa and I chatted for a while and a number of my comments about Alexa-type devices, but not all, were reported on air that evening. That story, which was a story about a story about Alexa, was rapidly syndicated and picked up around the world. Within a few days, the logo cloud of media sites that were quoting me looked a bit like this:
By the time I spoke with Dotun it was clear to me just how easy it was for someone to 'accidentally' buy something with an Amazon Echo. The magic word is not dollhouse, it could be drone or hoverboard; the "magic" is Alexa, which triggers a response from these devices. In the default configuration, the state of the system if you simply take it out of the box and plug it in following the installation instructions, is a. linked to your Prime account, and b. prepared to place orders with a simple verbal confirmation (using your "1-Click" settings as default payment method and shipping address).
And to be clear, Alexa will offer to ship you products even if you are not talking about buying something. For example, suppose you say, "Alexa, what's the best hoverborard?" The response will be a recitation of the product listing for the top rated hoverboard currently offered for sale on Amazon, immediately followed by an offer to ship it to you. If you say no, Alexa will then describe another product and offer to ship that. You need to say something like "Alexa cancel" or Alexa stop" to terminate the conversation. It so happens that the dollhouse ordered by the young girl that sparked the story was the second offering, suggesting that she had refused the first offer to send her a dollhouse.
At some point Amazon may enable two Echo features that could further reduce problems. First, allow owners of the devices to set a custom trigger word. At least that would enable you to talk about Alexa without waking her up. Second, but harder, would be to limit the voices to which Alexa responds, namely authorized users only. Of course, all of these things would add "friction" to the customer experience, which Amazon may be loath to do.
One question remains in my mind: Did Amazon ever consider that TV broadcasts would trigger the device? The CW6 experience was random, an accident. But if you intentionally broadcast the right words with the right timing you could trigger a mass ordering of products. And while Amazon has said it will accept returns of all 'accidental' orders, you can't use your Echo to cancel purchases. You have to go to the Amazon website or mobile app. Imagine a malicious broadcast that ordered expensive baby carriages, not the easiest things to return to sender. Does Amazon have an algorithm to detect that? Would some percentage of the orders be undetected until they turned up on doorsteps? How much would that cost in terms of dollars and good will?
And of course, buying things is not the only thing these devices can do. They can control thermostats and door locks and all manner of Internet of Things (IoT) devices. Pair that with the malicious broadcast scenario and you have some frightening possibilities. (I have been writing and talking about abuse of the IoT at We Live Security and other places.)
