Alexa… Help Me Solve this Murder

In a first of its kind case, police ask Amazon for data to help solve a murder.

UPDATE, 3/7/2017: Arkansas murder defendant James Bates, charged with murdering a man in his hot tub, has given police permission to review information that may be revealed in his smart home Echo recordings. Police hope to find aural evidence of the murder. Access to these routine or accidental recordings in response to a misheard wake word was twice denied by Amazon. The company repeatedly stated that its users' requests and their replies were protected by the First Amendment. Since permission was granted to access those records, the question of whether or not Echo/Alexa voice clips are protected by the First Amendment remains up in the air.

Original story continues:

Amazon has been asked to produce audio recordings from its personal assistant device, Alexa, in a murder trial. This is the first time such a request has been made. What legal and privacy issues arise from this request? What data do personal assistants record and retain? How long are those recordings that reach the cloud for translation kept? Do all personal assistants, such as Siri, Google and Cortana face similar legal issues?

The case that started it all
James Andrew Bates of Bentonville, Arkansas had a “guys' night in” gathering with food, football, beer and vodka shots on November 21, 2015. The group of four got together at Bates’ house, expecting to have a good time. According to those at the party, it was a good night. Bates and his friends allegedly hung out until 1 a.m., when Bates decided to go to bed leaving two of his buddies drinking in the hot tub.

In the morning, one of the individuals was found dead in the backyard. Bates called 911, and claimed to have found his friend face down in the hot tub when he opened his back door. Police stated the deceased man’s blood and injuries pointed to a struggle. They also noted an attempt to clean the scene, and determined the death was not an accident.

The deceased, 47-year-old Victor Parris Collins, was a former Georgia police officer. The water in the hot tub was red tinged, containing what looked like body fluids and blood. The victim had swollen and bruised lips, blood coming from his mouth and nose, a black eye and cut on his eyelid.

Detectives also noted the hot tub’s rim and the patio had been hosed off. Two cushions and three tub knobs were scattered on the ground, indicating a possible struggle or fight. There appeared to be diluted blood spots on the cover of one cushion, and there was also blood spatter on the sides of the hot tub and the hot tub cover. None of these observations would seem to indicate a drowning as alleged by Bates.

Phone records indicated Bates had made several calls, some to his parents and some to the other friend who he claimed had remained in the hot tub with the victim. In reality, that friend had been given a ride to his home just down the road, and had arrived at 12:30 a.m. The calls were made between 12:53 a.m. and 4:22 a.m. on November 22, 2015, when Bates was allegedly sleeping. Bates was the only person at the scene at the time of the death.

In line with detectives' suspicions, the Arkansas Chief Medical Examiner determined that Victor Collins died by strangulation, with drowning as a contributing cause, and ruled the case a homicide. The blood spatter samples were identified as belonging to Collins.

Police arrested and charged 31-year-old Bates with first-degree murder on February 22, three months after the incident.

The core issue in this case
The core issue in this case centers around the police requesting audio recordings from James Bates' personal assistant device, an Amazon Echo found on his premises: the same device that was streaming music near the hot tub where Collins was found dead. Detectives want to retrieve information to help support the homicide case against Bates.

This request to obtain information from a personal assistant is the first of its kind. One can certainly predict it is not likely the last request makers of personal assistant devices are going to receive.

The Echo is a speaker that also includes a personal assistant feature that reacts to voice commands. It only responds, in theory, to its individual “wake word.” In this case, the wake word would have been “Alexa.” Once activated, Alexa also records requests spoken to it. Those questions may be erased.

Evidently, some personal assistants also respond to words they perceive to be a request, which may not be intended as such. This means they are responding to a voice that may have said something that the internal software mistakenly recognized as its wake word. On this basis, police are requesting a warrant to review any activity that may have been recorded, intentionally or accidentally.

The issues surrounding personal assistants go deeper. Not only does the personal assistant sometimes mistakenly respond to what it deems is a wake word, it also responds to what it perceives to have been a question it was not asked — meaning it is picking out random words or parts of sentences, interpreting them as a request and offering a response based on what it mistakenly perceived. It then becomes clear why the police would want to access the audio logs from the Echo located next to the hot tub.

Colorado criminal defense attorney, Miller Leonard describes the situation by saying, “The probable cause necessary to search a computer or electronic media is if there is probable cause to believe that the item in question contains or is contraband, or contains evidence of a crime, or fruits of a crime, or instrumentalities of a crime. Evidence of crime can include evidence of ownership and control. According to the Supreme Court, the probable cause standard is satisfied by an affidavit that establishes ‘a fair probability that contraband or evidence of a crime will be found in a particular place.’ Thus, in the case in question, the prosecutor must have convinced a Judge that such probability existed.”

Some audio may have been retained on the Echo itself. Utterances are recorded by a microphone to a binary audio file and sent in a JSON (JavaScript Object Notation) message to the Alexa server. A JSON message is a data-interchange format, easy for machines to generate and easy for humans to read and write.

If nothing overwrote/erased the existing audio buffer files in Bates' Echo before it was taken by police as evidence, it may contain audio files from the night of the murder. Forensically speaking, police may also be able to recover any deleted audio from the local file system, much like retrieving deleted material from the hard drive on a computer.

Amazon initially refused to cooperate, but offered account information and a purchase history for Bates. They later provided partial information to the police in response to their warrants.

Detectives were also interested in accessing all other logs/records from any internet connected device present in Bates house, such as his water meter. Those logs revealed he had used over 140 gallons of water between the hours of 1 a.m. and 3 a.m., the night Victor Collins died, which supported the theory Bates had hosed the hot tub and patio down to hide evidence.

The legal questions surrounding the police's ability to access to personal information about an individual from internet-connected devices are still untested. On one side of the debate is the idea that using such personal information is an invasion of privacy. People who are using devices intended to enhance quality of life should still have a reasonable expectation of privacy in their own homes. Those on the other side of the argument say that if people with these devices are committing illegal acts, and those acts are recorded, police, judges and the public have a right to that information as evidence.

Always listening?
There is a distinction to be made between passive and active listening with personal assistants. It is not correct to say that all devices with microphone-enabled capabilities are always on. In fact, many devices referred to as always on are only using the microphone to detect a wake word or phrase, meaning something has to trigger them.

What is more accurate is that there are three categories these devices may be placed into, allowing for the fact that some devices do more than just one thing. Those three categories are:

• Manually activated devices that require a user to perform a physical action to turn the device on
• Speech activated devices that require a wake word or phrase
• Always on devices that constantly transmit data, like security cameras

Each category has different privacy concerns based on whether or not the data obtained from them is stored externally in the Cloud, by a third party or stored locally. It is also critical to know whether or not the device in question is used for voice recognition (biometric identification) or speech recognition, as in translating voice to text.

Speech-activated devices
The Echo is a speech-activated device that relies on processors to remain in a state of passive processing and listening for a wake word. It buffers and re-records locally but does not transmit or store any information until it hears its wake word. When the wake word is used it starts recording.

The Echo then is not really listening to its environment; it is using the microphone as another environmental sensor. This distinction will be crucial in lawsuits alleging invasion of privacy.

Always on devices
Always on devices are designed to record and send information constantly. Many Americans have these devices in the form of baby monitors, home security cameras, or newer devices like the OrCam 13s, a wearable video cam for those who are visually impaired that translates text to audio in real time.

Regardless of the method used to activate a device, legal issues relating to the status of voice data are bound to arise. This is new territory that many current laws were not written to address. In some cases the authors could not have anticipated such technology. Current legislation that may apply relates to laws that protect biometric information. Other regulations and laws may apply based on the content of the voice communications.

Lawyers will have to wait and see how courts treat requests for data from personal assistants and whether such requests prompt new legislation. Currently, police may attain a warrant for cellphone data when investigating accidents or other crimes in which the victim or perpetrator carried a smartphone. It is not much of a stretch to imagine personal assistants being added to the arsenal of crime detection tools available to law enforcement.