SAN FRANCISCO - Tesla has assured its millions of electric car owners that their privacy “is and will always be enormously important to us”.

The cameras it builds into vehicles to assist driving, it notes on its website, are “designed from the ground up to protect your privacy”.

But between 2019 and 2022, groups of Tesla employees privately shared, via an internal messaging system, sometimes highly invasive videos and images recorded by customers’ car cameras, according to interviews by Reuters with nine former employees.

Some of the recordings caught Tesla customers in embarrassing situations. One former employee described a video of a man approaching a vehicle completely naked.

Also shared were crashes and road-rage incidents. One crash video in 2021 showed a Tesla being driven at high speed in a residential area and hitting a child riding a bike, according to another former employee.

Other images were more mundane, such as pictures of dogs and funny road signs that employees made into memes by embellishing them with amusing captions or commentary, before posting them in private group chats. While some postings were shared between only two employees, others could be seen by scores of them, according to several former employees.

Tesla states in its online “Customer Privacy Notice” that its “camera recordings remain anonymous and are not linked to you or your vehicle”. But seven former employees told Reuters the computer programme they used at work could show the location of recordings – which potentially could reveal where a Tesla owner lived.

One former employee also said that some recordings appeared to have been made when cars were parked and turned off. Several years ago, Tesla would receive video recordings from its vehicles even when they were turned off, if owners gave consent. It has since stopped doing so.

“We could see inside people’s garages and their private properties,” said another former employee. “Let’s say that a Tesla customer had something in their garage that was distinctive, you know, people would post those kinds of things.”

Tesla did not respond to detailed questions sent to the company for this report.

To report this story, Reuters contacted more than 300 former Tesla employees who had worked at the company over the past nine years and were involved in developing its self-driving system.

More than a dozen agreed to answer questions, all speaking on condition of anonymity.

Reuters was not able to obtain any of the shared videos or images, which former employees said they had not kept. The news agency also was not able to determine if the practice of sharing recordings, which occurred within some parts of Tesla as recently as 2022, continues today or how widespread it was.

Some former employees contacted said the only sharing they observed was for legitimate work purposes, such as seeking assistance from colleagues or supervisors.

The sharing of sensitive videos illustrates one of the less-noted features of artificial intelligence systems: They often require armies of human beings to help train machines to learn automated tasks such as driving.

Since about 2016, Tesla has employed hundreds of people in Africa and later the United States to label images to help its cars learn how to recognise pedestrians, street signs, construction vehicles, garage doors and other objects encountered on the road or at customers’ houses. To accomplish that, data labellers were given access to thousands of videos or images recorded by car cameras that they would view to identify objects.

Two former employees said they were not bothered by the sharing of images, saying that customers had given their consent or that people long ago had given up any reasonable expectation of keeping personal data private. Three others, however, said they were troubled by it.

One said: “I’m bothered by it because the people who buy the car, I don’t think they know that their privacy is, like, not respected... We could see them doing laundry and really intimate things. We could see their kids.”

One former employee saw nothing wrong with sharing images, but described a function that allowed data labellers to view the location of recordings on Google Maps as a “massive invasion of privacy”.

Associate Professor David Choffnes, executive director of the Cybersecurity and Privacy Institute at Northeastern University in Boston, called the sharing of sensitive videos and images by Tesla employees “morally reprehensible”.

He noted that circulating sensitive and personal content could be construed as a violation of Tesla’s own privacy policy – potentially resulting in intervention by the US Federal Trade Commission (FTC), which enforces federal laws relating to consumers’ privacy.

An FTC spokesman said it does not comment on individual companies or their conduct.

To develop self-driving car technology, Tesla collects a vast trove of data from its global fleet of several million vehicles. The company requires car owners to grant permission on the cars’ touchscreens before Tesla collects their vehicles’ data. “Your data belongs to you,” states Tesla’s website.

In its customer privacy notice, Tesla explains that if a customer agrees to share data, “your vehicle may collect the data and make it available to Tesla for analysis. This analysis helps Tesla improve its products, features, and diagnose problems quicker.”

It also states that the data may include “short video clips or images”, but is not linked to a customer’s account or vehicle identification number, “and does not identify you personally”.

Mr Carlo Piltz, a data privacy lawyer in Germany, told Reuters it would be difficult to find a legal justification under Europe’s data protection and privacy law for vehicle recordings to be circulated internally when it has “nothing to do with the provision of a safe or secure car or the functionality” of Tesla’s self-driving system.

Tesla calls its automated driving system Autopilot.

Introduced in 2015, the system included such advanced features as allowing drivers to change lanes by tapping a turn signal and parallel parking on command.

To make the system work, Tesla initially installed sonar sensors, radar and a single front-facing camera at the top of the windscreen.

A subsequent version, introduced in 2016, included eight cameras all around the car to collect more data and offer more capabilities.

Tesla chief executive Elon Musk’s future vision is eventually to offer a “Full Self-Driving” mode that would replace a human driver. Tesla began rolling out an experimental version of that mode in October 2020.

Although it requires drivers to keep their hands on the wheel, it currently offers such features as the ability to slow a car down automatically when it approaches stop signs or traffic lights.

To develop Autopilot, Tesla hired data labellers to identify objects in images and videos to teach the system how to respond when the vehicle was on the road or parked.

Tesla initially outsourced data labelling to a San Francisco-based non-profit then known as Samasource, people familiar with the matter told Reuters. In 2016, Samasource was providing about 400 workers in Kenya for Tesla, up from about an initial 20, according to a person familiar with the matter.

By 2019, however, Tesla was no longer satisfied with the work of Samasource’s data labellers.

A former Tesla employee said of the Samasource labellers: “They would highlight fire hydrants as pedestrians... They would miss objects all the time. Their skill level to draw boxes was very low.”

Tesla then brought data labelling in-house. “Over time, we’ve grown to more than a 1,000-person data labelling (organisation) that is full of professional labellers who are working very closely with the engineers,” Tesla’s then senior director of AI Andrej Karpathy said in an August 2021 presentation.

Tesla’s own data labellers initially worked in the San Francisco Bay Area, including the office in San Mateo.

Groups of data labellers were assigned a variety of different tasks, including labelling street lane lines or emergency vehicles, former employees said.

At one point, Teslas on Autopilot were having difficulty backing out of garages and would get confused when encountering shadows or objects such as garden hoses. So some data labellers were asked to identify objects in videos recorded inside garages. The problem was eventually solved.

In interviews, two former employees said in their normal work duties that they were sometimes asked to view images of customers in and around their homes, including inside garages.

“I sometimes wondered if these people know that we’re seeing that,” said one.

“I saw some scandalous stuff sometimes, you know, like I did see scenes of intimacy but not nudity,” said another. “And there was just a lot of stuff that I wouldn’t want anybody to see about my life.”

As an example, this person recalled seeing “embarrassing objects”, such as “certain pieces of laundry, certain sexual wellness items... and just private scenes of life that we really were privy to because the car was charging”.

Tesla staffed its San Mateo office with mostly young workers in their 20s and early 30s, who brought with them a culture that prized entertaining memes and viral online content.

Former staff described a freewheeling atmosphere in chatrooms, with workers exchanging jokes about images they viewed while labelling.

According to several former employees, some labellers shared screenshots, sometimes marked up using Adobe Photoshop, in private group chats on Mattermost, Tesla’s internal messaging system. Participants would add their own marked-up images, jokes or emojis to keep the conversation going. Some of the emojis were custom-created to reference office inside jokes.

One former labeller described sharing images as a way to “break the monotony”. Another described how the sharing won admiration from peers.

“If you saw something cool that would get a reaction, you post it, right, and then later, on break, people would come up to you and say, ‘Oh, I saw what you posted. That was funny’,” said this former labeller. “People who got promoted to lead positions shared a lot of these funny items and gained notoriety for being funny.”

There were dogs, interesting cars, and clips of people recorded by Tesla cameras tripping and falling. There was also disturbing content, such as someone being dragged into a car seemingly against their will, said one former employee.

Video clips of crashes involving Teslas were also sometimes shared in private chats on Mattermost, several former employees said. Those included examples of people driving badly or collisions involving people struck while riding bikes or a motorcycle.

Some data labellers would rewind such clips and play them in slow motion.

At times, Tesla managers would crack down on inappropriate sharing of images on public Mattermost channels since they claimed the practice violated company policy. Still, screenshots and memes based on them continued to circulate through private chats on the platform, several former employees said.

One of the perks of working for Tesla as a data labeller in San Mateo was the chance to win a prize – use of a company car for a day or two, said two former employees.

But some of the lucky winners became paranoid when driving the electric cars.

“Knowing how much data those vehicles are capable of collecting definitely made folks nervous,” one former employee said. REUTERS