Hacked autonomous vehicles could cripple New York City
Two research teams highlight individual and mass dangers of vulnerable vehicles. Richard A Lovett reports.
Scientists concerned about the possibility of large-scale hacking attacks on autonomous cars are using mathematics originally developed for modeling the percolation of water through coffee grounds to determine how such an attack might affect traffic in congested areas like New York City.
The goal, says Skanda Vivek, a physicist at Georgia Institute of Technology, Atlanta, who presented his results at a meeting of the American Physical Society (APS) in Boston, Massachusetts, isn’t to frighten people; it’s to learn how to protect cars of the future from hacking assaults before they occur.
In 2015, Vivek says, a reporter from Wired magazine collaborated with a pair of hackers to prove that an Internet-connected car – in that case a Jeep Cherokee – could be hijacked by hackers operating from kilometres away. And that wasn’t even a fully autonomous car – it was a fairly normal automobile with Internet-connected features the hackers used to override the driver’s control.
Such a hack takes advantage of the way information flows within a vehicle, says J Christian Gerdes, director of the Centre for Automotive Research (CARS) at Stanford University, Palo Alto, who is not a member of Vivek’s team.
“For instance, the car’s computer needs to communicate commands to the engine/motor, steering, and brakes,” he says. In this case, he adds, the Wired hack came through the car’s infotainment system, which was on the same network as the vehicle’s internal controls.
A more fully automated vehicle might be hacked, Gerdes says, by inserting malware into its software upgrades or by compromising its sensors so that they misunderstand their surroundings. It might also be possible to alter an autonomous car’s processing systems so they mischaracterise things they see, thereby making wrong decisions.
Such attacks could be used for any of a number of purposes, ranging from demands that owners pay a ransom in order to regain control of the vehicle to personal vendettas, or even attempted murder by ramming the car into a bridge abutment.
“Companies are aware of this possibility and guard against it,” Gerdes says, “but it highlights the importance of thinking through all of the information flows.”
Vivek’s interest is broader: the use of coordinated attacks to bring down entire traffic grids, as either a hack or – a term he avoids – terrorism.
Under this scenario, some cars might react by losing touch with their environments and colliding with things they can no longer see, such as the vehicle in front of them. Others might go into a failsafe mode and stop suddenly in the middle of the road. Either way, Vivek says, they become obstacles to traffic flow.
And that’s where coffee grounds come into play.
When you’re making your morning cup of java, what you want is for the hot water to percolate through the coffee grounds, not too fast, not too slow. The maths of percolation theory, which can apply to other problems involving fluids and small grains, are what dictate how this happens.
For the types of total traffic gridlock Vivek’s team is worried about, the maths are the same, but the outcome is the reverse. As more and more cars are affected, there comes a point where nothing can percolate through the blockages created until tow trucks come in from unaffected areas and laboriously clear away the stalled or crashed vehicles, one by one.
But however badly an accident or stalled vehicle might mess up the morning commute, it takes more than a few stalls to totally shut down a city and make it impossible for anything – including emergency vehicles – to get anywhere.
In the work also presented at the APS conference, Vivek’s team modelled randomly distributed stalls or crashes on Manhattan Island and found that if somewhere around 10% of all vehicles were disabled, the island’s 1.7 million residents would wind up totally cut off from the rest of New York.
“Essentially, vehicles would be blocked in such a way that you can’t pass through,” he says.
Lower rates of disablement would also be extremely disruptive, but there’s a difference between a wickedly slow commute and not being able to get anywhere at all.
At the moment, of course, far fewer than 10% of automobiles are sufficiently wired to make a large-scale hack possible. But as self-driving cars become more common, Vivek says, it’s important to think about how to take precautions.
For example, each of the top four auto manufacturers selling cars to New Yorkers each has around 10% of the market. Thus, if all the cars made by any one of them were vulnerable to an attack, that would be enough for a catastrophe.
One safety precaution, Vivek explains, is for manufacturers to make sure that all of their cars, or even all of the cars of the same model, don’t use the same software.
“Make it hard to hack so many at the same time,” he says, “so you can only hack 2%, not 10%.”
That said, Vivek doesn’t think there’s any reason to be overly fearful of Internet-connected cars.
“[These] cars are the future,” he says. “There’s a lot of potential benefits, including reducing traffic congestion and reduced fuel consumption that could be highly beneficial to the environment.”
Instead, he adds, “by shining a light on these technologies at an early stage, we hope we can help prevent worst-case scenarios”.