In order to turn highly automated and autonomous driving into a broad scenario suitable for everyday use, car manufacturers and equipment suppliers still have to make considerable efforts, especially in vehicle safety. This was emphasized by experts on Thursday at the online conference “SOS Secure Our Streets” organized by the non-profit Automotive Security Research Group, which is dedicated to promoting and developing security solutions for the automotive industry.
“Radical changes with significant investments”
In principle, the industry must redundantly create the entire complex IT system that makes up a robot car with detection functions such as cameras, radar and lidar, explained Cristian Ion, who is responsible for the IT security company Cymotive, founded in 2016 by Volkswagen and Israeli security experts engineering is responsible. A considerable increase in computing and communication capacities is necessary in order to enable “high throughput of cryptographically authenticated data”.
It is just as necessary to guarantee external data quality and ensure secure fleet operations centers and secure development and operational processes, explained Ion. “All of these areas represent a serious challenge in their own right.” Taken together, making the dream of the self-driving car a reality required “a radical change in architecture with significant investment.”
Dangers – also through “Tiktok Challenges”
The industry shouldn’t just count on the hacker next door, the expert warned. Rather, the profile of the attackers must include “Tiktok challenges” as well as “naval sabotage, political activism and national actors”. The previous categories for claims no longer applied if, at a speed of 130 km/h, there was only a few seconds in an emergency for the vehicle to be taken over by a person or for the vehicle to be pulled out of traffic in a controlled manner.
According to Ion, there are also cracking nuts such as jamming radar or communication functions, reading out manipulated QR codes or precise location. The artificial intelligence (AI) systems used have also not yet proven to be secure and robust. All difficulties, such as the authentic and correct analysis of data that would also come from other connected vehicles and the infrastructure, would have to be solved in real time. The most recent taxi traffic jam in Moscow after a hack by the dispatching company and an hour-long blockade in San Francisco by self-driving cruise cars gave an idea of the disaster that threatened the fleet operation.
Traditional security approach not practical
“Some suggest a secure remote control,” the technician referred to a potential handle. The inclusion of appropriate tele-operators is required by law in this country. However, this requires secure operating centers with protected remote control processes and a robust service ecosystem, emphasized Ion. The integrity of the software supply chain must also be guaranteed.
Tao Zhang, who heads a working group for transformative networks at the US standardization institute NIST (National Institute of Standards and Technology), also does not consider the traditional security approach for autonomous vehicles to be practicable. For example, not every engine control and other IT unit in a car can be shielded with a firewall. The method of switching off, cleaning and restarting an attacked system doesn’t work with moving subsets either. Software updates alone are also insufficient, but car hardware can only be replaced in individual cases. Not to mention privacy issues.
Edge computing as a possible solution
Zhang is hoping for a solution from edge computing, i.e. the massive processing of large amounts of data from networked cars and the road infrastructure as close as possible to where it is happening in small, ideally portable data centers at the edge of the network. It is hardly possible anymore to transmit all relevant measured values via the cloud to large, more distant data centers and only evaluate them there. However, the local nodes could become smarter via the computer clouds through “federated learning”: they only exchanged the extracted know-how, such as new AI training programs.
This is how a distributed firewall could be set up, the NIST representative gave an example. The cars as well as the edge and cloud centers should train AI models together. At the same time, some raw data could be exchanged “strategically in tiny portions” as long as control over it was given. In general, it is important to ensure that the selected AI models are transparent and not compromised. In order not to put too much strain on the batteries, especially in electric vehicles, the computing resources in the car itself would have to be used very carefully.
Commercial vehicles: “The hackers laugh up their sleeves”
Gilad Bandel from the business development department at Cymotive has already identified commercial vehicles without autonomous functions as particularly vulnerable to cyber attacks: almost all trucks use the SAE J1939 network protocol with only slight variations in implementation, which allows communication on a CAN bus (Controller Area Network) for the transmission of diagnostic data and control information. If a vulnerability were found in it, it could be exploited “over and over again” in all makes and models of different outfitters and vehicle manufacturers.
Unfortunately, J1939 was designed a little older and “without cyber security,” Bandel reported: “The hackers are laughing up their sleeves.” Thanks to component interoperability, they would only have to find a single successful line of attack. There is no authentication on the bus itself, so messages could easily be forged to carry out a man-in-the-middle attack. Even a mediocre hacker could develop an “end-to-end attack” in a week or two.
Recommended Editorial Content
With your consent, an external survey (Opinary GmbH) will be loaded here.
Always load polls Load poll now
In addition, according to the expert, fleet managers often add devices and updates afterwards, which increases the attack surface. For example, some installed Electronic Logging Devices (ELDs) and other telematics equipment to monitor drivers. These are not part of the supply chain and are particularly easy to compromise. Other integrated software often contains many vulnerabilities. A SA J1939-91 protocol extension promises a remedy for network security, which also allows secure updates via the air interface. But this is still in the works. Until then, special intrusion detection systems (IDS) could alleviate the problems. However, they cannot be installed in commercial vehicles via “Plug & Play”.
To home page