IEEE Consumer Electronics Magazine - November/December 2022 - 16

Securing the Future of Transporation
Another recent attack thatmade the headlineswas
the T-BONE attack12 where researchers were able
to gain remote code execution over WiFi on the
infotainment system in a Tesla Model 3 using a
drone. They were able to remotely open doors and
trunk, change seat positions, steering, and acceleration
modes. However, this exploit does not provide
driving control of the vehicle. The
researchers also highlighted that adding a privilege
escalation exploit, such as CVE-2021-3347 to TBONE,
would weaponize this exploit and turn it
into a worm. This would allow them to load new
WiFi firmware and exploit other Tesla cars in the
victim car's proximity. More recently, in the beginning
of 2021, an online hacking group by the name
DoppelPaymer claimed to have conducted a ransomware
attack on KIA motors America and have
stolen unencrypted confidential data.13
Thus, cyber-attacks are becoming increasingly
prevalent in modern vehicles. A comprehensive
taxonomy of vehicular security attacks
with details related to attacker types, attack
tools and actions, and attack objectives is discussed
by Thapliyal et al.31
AI TO THE RESCUE FOR INTRUSION
DETECTION IN VEHICLES
To protect vehicles from devastating cyberattacks,
it is crucial to address cybersecurity in
the automotive domain. Several lightweight
authentication mechanisms, such as the works
of Wang and Sawhney25 and Labrado et al.,33 and
versatile automotive security frameworks, such
as Kukkala et al.'s work,26 try to improve security
in resource-constrained automotive ECUs. However,
these techniques mainly focus on authenticating
the ECUs during the initial setup phase
and cannot detect the presence of an attacker in
the later stages or when an already authenticated
ECU gets compromised. Thus, there is a
need for an IDS that continuously monitors the
vehicular network and detects any cyberattacks.
Moreover, an IDS is extremely crucial
and often the last line of defense when the
attacker breaks through defense mechanisms.
Traditional IDS solutions relied on firewalls or
rule-based (non-AI-based) systems to detect
cyber-attacks. However, they are not effective
against detecting sophisticated automotive
16
attacks as they fail to capture the complex
dependencies in the time series vehicular network
data. With the availability of large troves of
data in vehicles from communication between
ECUs and external systems, and the increased
computing capabilities of ECUs, AI-based solutions
can be leveraged to parse high-dimensional
vehicular network data. Here, we discuss some
recent work on AI-based IDSs that encompass
monitoring and attack detection at in-vehicle
network level and vehicular ad hoc network
(VANET) level.
IDSs for In-Vehicle Network Security
A generative adversarial neural network
(GAN) based IDS was introduced by Seo et al.14
that tries to learn the patterns of the message
identifier (ID) in CAN data, by transforming the
ID data into " CAN images, " to detect attacks.
This approach employs two discriminator models
to detect known and unknown attacks,
respectively. The former is trained using normal
and abnormal CAN images from actual vehicle
data, and the latter is trained simultaneously
using the adversarial process (a mix of adversarial
data from the generator and normal CAN
images). The second discriminator learns to
detect fake CAN images that look similar to real
CAN images. This approach showed high detection
accuracy against distributed denial of service
(DDoS), and message injection attacks.
A gated recurrent unit (GRU) based recurrent
autoencoder IDS is presented by Kukkala et al.18
During training, this technique learns the normal
operating behavior of the system by reconstructing
the input attack-free CAN data. During inference
at runtime, this technique observes the
deviation from the learned normal system behavior
to detect cyber-attacks. This approach was
implemented on real-world automotive hardware
and tested under several attack scenarios. Kukkala
et al.16 proposed an unsupervised LSTM-based
encoder-decoder that integrates a novel self-attention
mechanism to learn the characteristics of
normal data traversing the in-vehicle network. The
attention mechanism enhanced the ability of the
model to focus on the important information from
the past. A one-class support vector machine
(OCSVM) based classifier was trained and
deployed together with the self-attention model at
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - November/December 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - November/December 2022
