Wake up to Mike Allen’s AM, the 10 stories driving your day

Stories

Expert Voices

Machine learning could protect AV systems from cyberattacks

Illustration of car surrounded by artificial intelligence dome
Illustration: Sarah Grillo/Axios

With cars further evolving from “just” hardware into software platforms, the wide range of IT systems in an autonomous vehicle present many opportunities for cyberattacks. Artificial intelligence, especially machine learning technologies, can detect anomalous vehicle behavior or attempted code interference in real time.

Why it matters: Compromised code in the systems that control acceleration, braking, directional guidance and other crucial safety functions can jeopardize lives on the road.

Inside a hack:

  • After hackers breach an AV’s system, they can identify vulnerabilities, infect additional areas of the vehicle, and collect and extract data.
  • Their malicious goals could include anything from tracking the vehicle's movements (perhaps with an eye toward robbing passengers or burglarizing their homes when they are far enough away) to disabling its brakes, putting people both inside and around the vehicle in danger.
  • Embedded, stealth malware in the vehicle's network and in the units that control its IT and electronic subsystems could generate atypical activity, such as fluctuations in the number of bytes sent or the amount of memory being used.
  • Example: Hackers could install malware in a truck that activates when it determines — via the vehicle's cameras and sensors — the truck is in an isolated section of highway. The truck could be forced to stop, allowing criminals in another vehicle to corner it and abscond with the loot. 

How AI detection might help:

  • AI uses data from the vehicle's network and central computers to learn its normal behavior, typically before the vehicle is mass-produced. Algorithms can then process operating data to identify and flag any abnormal behavior as a potential cyberattack, triggering safety systems that can intervene.
  • These algorithms need to distinguish between real attacks and false positives, such as when a vehicle owner has modified the software for speed or other safety precautions.

What to watch: Attacks on AVs could spur a “horse race” between developers trying to continually improve traditional security measures and hackers developing more sophisticated attacks to upend these security systems.

Yossi Vardi is the CEO of SafeRide Technologies, an automotive cybersecurity startup.

More stories loading.