McAfee warns of autonomous car vulnerability
Software security researchers apply sticker to speed limit sign for "targeted misclassification"Researchers at McAfee Advanced Threat Research have found a 'vulnerability' in the mobilEye Q3 camera technology of the Tesla Model S.
By simply applying a piece of black electric tape to a 35MPH (56km/h) sign, the researchers fooled the on-board camera to read the sign as an 85MPH (137km/h) speed limit sign. And for this video demonstration, in what the authors of a McAfee study describe as a case of "adversarial machine learning" or "model hacking", the Tesla did indeed commence accelerating to 85MPH. The driver overrode the program at 50MPH (80km/h) out of safety concerns.
According to the study's authors, Steve Povolny and Shivangee Trivedi, the latest Model S with an upgraded mobilEye camera isn't so easily fooled, but a Model S (pictured) or Model X from 2016 with 'Hardware Pack 1' and relying on ' Tesla Automatic Cruise Control' (TACC) is susceptible. The TACC adjusts the car's speed automatically in accordance with the prevailing speed limit sign.
Obviously the demonstration video highlights just one type of vulnerability, and though Tesla and mobilEye may have addressed that vulnerability, that doesn't mean other car companies and parts suppliers have. According to McAfee, mobilEye has supplied cameras for over 40 million vehicles around the world.
The authors of the study say that most vehicles built in 2019 have "state-of-the-art sensors" that could be deceived. These devices include rain sensors for wipers, radar for AEB and presumably ultrasonic sensors for acoustic parking guidance. All could be potentially susceptible to 'adversarial machine learning'.
To date, the authors say, there has been no evidence of black-hat hackers carrying out this sort of attack on an autonomous or semi-autonomous vehicle, and McAfee ATR pats itself on the back for beating the "adversaries in identifying weaknesses in underlying systems".
While the research body has a vested interest in highlighting vulnerabilities, it has done the right thing, providing both Tesla and mobilEye with its findings up to 90 days ahead of public disclosure, allowing the two companies time to redress any deficiencies. According to McAfee, mobilEye has revealed to the research team that its latest on-board systems do address these vulnerabilities.
The McAfee ATR study is the result of research over 18 months, which trialled the use of other stickers to fool the on-board camera. A stop sign was perceived to be an 'added lane' sign after modification with tape. At the present time cars do not automatically stop for stop signs, but it's only a matter of time before that function is embraced by car companies adding to the complement of autonomous driving features in the modern car.
As a result of its work, McAfee ATR has developed its own 'classifier', which is not easily deceived by the trickery of defacing road signs with stickers. In the case of the modified 35MPH speed limit sign, the McAfee classifier classified the sign with 95.93 per cent "confidence".
While the study has proved useful to McAfee in determining some of the flaws in current autonomous motoring technology, the real take-out for the general public is that you don't need to be a high-tech hacker to hoodwink the automotive world's smartest (silicon) brains.
You just need some black friction tape.
Please see our Editorial Guidelines & Code of Ethics (including for more information about sponsored content and paid events). The information published on this website is of a general nature only and doesn’t consider your particular circumstances or needs.
