Miller and Valasek tore apart the dashboards of the vehicles to learn how the various automated features were networked and run, and ultimately wrote code to control the electronics that run the steering wheel, brakes, and other functions. Their work follows the remote car-hacking research in 2011 by the University of Washington and the University of California-San Diego, which found ways to hack car features via Bluetooth and rogue CDs, for instance. The academics kept private some details of the hacks, including which cars they were able to "own."
So Miller and Valasek wanted to take car hacking a step further and drill down and see what really could be done if a hacker got inside the car's network. They also wanted to share those details, as well as the car-hacking tools they built, with other researchers.
"This addresses a problem that doesn't really exist yet but we're afraid will, very soon. We want to get ahead of the curve," Miller, who is a security engineer at Twitter, said in an interview with Dark Reading.
The pair, known for their software-hacking, spent about 10 months working on the hacks. Miller says they hope the tools will help other researchers find other weaknesses in vehicles that should be addressed.
Miller says he and Valasek hope to also help create a "car-in-a-box" research model that simulates the vehicles so you don't need to spend $40,000 on a vehicle to test.
"One of our big goals is to get other people involved in this. But obviously, there's a scale issue ... people don't have [money] to buy the cars," he says. "We want to lower that barrier to entry, with a car-in-a-box [model] ... where you could do simulation and research on this that you know would work in a real car."
[Researcher demonstrates how 'horrifyingly' easy it is to disarm a car alarm system and control other GSM and cell-connected devices. See 'War Texting' Attack Hacks Car Alarm System .]
At the heart of their car-hacking research was cheating the Controller Area Network (CAN) technology resident in the vehicles by injecting their own messages that would disable the Ford Escape's breaks at slow speeds or kill its engine while on the road, for example. They were able to force the Toyota Prius to brake at 80 miles per hour and to jerk the steering wheel out of the driver's control.
CAN supports the electronic control units (ECUs) for each of the car's automated features. "Almost of the [automated] features in the Prius are controlled by computers that take input from sensors," Valasek told DEF CON attendees. "The ECUs are connected by the CAN bus ... once you're on the CAN bus ... and if you compromise an ECU or are connected in physically, you're all good."
The researchers sent their own messages to the CAN network in the hacks. In one hack captured on videotape, the researchers simulated a message that the Prius's gas pedal was "floored" when it was not. The car ultimately stopped, but the engine was still revving up as if the accelerator was engaged. "This was done by the injection of 'normal' CAN traffic," said Valasek, who is director of security intelligence for IOActive.
Miller and Valasek's tools include EcomCat, a "Swiss Army knife" that can read and write to the CAN bus, store output from the CAN bus, and read and write from file, for example; EcomCat API, an application programming interface for EcomCat; and PyEcom, a Python tool.
But not all of the cars' functions run over CAN. The directly wired functions like the speedometer, lights, and horn can be manipulated by injection attacks on the diagnostics, the researchers found. "You can take out the ECU in control of all of the lights in the car. So you can turn the lights out, and the AC goes out, the radio goes out, the brake lights go out, and you can't get out of park, either," Miller told DEF CON attendees.
Toyota and Ford have publicly played down the research. The researchers provided both auto companies with their white paper, but neither firm has promised any fixes. Ford said in a statement for a "Today" show segment featuring Miller and Valasek's research that "This particular attack was not performed remotely over-the-air, but as a highly aggressive direct physical manipulation of one vehicle ... which would not be a risk to customers."
Miller says Toyota's response was similar, saying its focus was on remote attacks and that this research didn't constitute hacking. It's unlikely that Ford or Toyota will address the security issues in the end, he says. "It's not enough to get them to do anything. And it's not a Web browser bug, where you could release a patch. It's going to be difficult" to secure these systems in the cars, he told Dark Reading.
There are some steps the carmakers could make, though, according to Miller, such as isolating CANs and adding a detection-type system on the network to catch any suspicious or unauthorized behavior.
Miller and Valasek's white paper and tools can be downloaded here.
Have a comment on this story? Please click "Add Your Comment" below. If you'd like to contact Dark Reading's editors directly, send us a message.