Tesla’s autopilot may be less reliable than the electric car maker claims. With a small mark made with a piece of tape, a group of researchers has managed to trick the company’s first-generation autopilot into varying its speed considerably.
Hackers have made a modification to a speed limitation road sign. The car was supposed to go below 35 miles per hour (that is, about 56 kilometers per hour) , however they have added a small sticker to make the 3 look like an 8 .
The Tesla has succumbed to the deception. The electric car’s artificial intelligence system has interpreted the speed limit sign to read 85 miles per hour (almost 137 kilometers per hour) instead. This subtle change has caused the vehicle to speed up to more than double the speed allowed on that road, posing a great danger to other road users.
In the eyes of any human driver, it can be clearly identified that, despite the piece of tape, it is still a 3 with some kind of addition, however, the Tesla AI has not been so clear.
In reality, it has not ‘stung’ a single Tesla, but two. The models that have been put to the test have been a Tesla Model X and a Model S , both from 2016. Both incorporate the Mobileye EyeQ3 camera system, capable of reading signals and sending the information to the car’s speed assistant. Tesla launched its Speed Assist in 2014 and has been improving it over time and testing.
Trolling Tesla for a good cause
The test for which he has ‘fooled’ Elon Musk’s car has come from an initiative of a group of hackers from the McAfee security company, specifically the McAfee Advanced Threat Research laboratory.
In a post they clarify what their intention was and what their goals are: “Our specific objective at McAfee Advanced Threat Research is to identify and shed light on a wide spectrum of threats in today’s complex landscape. With model hacking we study how enemies can attack and evade artificial intelligence. So we have an incredible opportunity to influence awareness, understanding and the development of safer technologies before they are implemented in a way that has real value to the enemy ”, they say.
The company’s security lab warned both Tesla and Mobileye of what happened to improve their technologies. He states that they have appreciated the investigation, although at the moment “they have not shared any current plan to address the problem on the existing platform.” From Mobileye they are sure that the problem would already be avoided with the most recent versions that have been introduced in their camera system.
What they say from McAfee is not nonsense. The experiment has revealed what could happen if malicious people with certain technical knowledge decide to act to cause one or more traffic accidents and the ease with which it is possible to fool the AI -even if it is from the past- of some cars with automatisms.
