Brain-Computer Interface Applications: Security and Privacy Challenges

Abstract

Brain-Computer Interfaces (BCI) are becoming increasingly popular in medical and non-medical areas. Unfortunately , manufacturers of BCI devices focus on application development, without paying much attention to security and privacy related issues. Indeed, an increasing number of attacks to BCI applications underline the existence of such issues. For example, malicious developers of third-party applications could extract private information of users. In this paper, we focus on security and privacy of BCI applications. In particular, we classify BCI applications into four usage scenarios: 1) neuromedical applications, 2) user authentication, 3) gaming and entertainment, and 4) smartphone-based applications. For each usage scenario, we discuss security and privacy issues and possible countermeasures.

Full text

Brain-Computer Interface Applications:
Security and Privacy Challenges
QianQian Li
University of Padua, Italy
qianqian@math.unipd.it
Ding Ding
University of Padua, Italy
ding@math.unipd.it
Mauro Conti
University of Padua, Italy
conti@math.unipd.it
Abstract—Brain-Computer Interfaces (BCI) are becoming in-
creasingly popular in medical and non-medical areas. Unfor-
tunately, manufacturers of BCI devices focus on application
development, without paying much attention to security and
privacy related issues. Indeed, an increasing number of attacks
to BCI applications underline the existence of such issues. For
example, malicious developers of third-party applications could
extract private information of users.
In this paper, we focus on security and privacy of BCI applica-
tions. In particular, we classify BCI applications into four usage
scenarios: 1) neuromedical applications, 2) user authentication, 3)
gaming and entertainment, and 4) smartphone-based applications.
For each usage scenario, we discuss security and privacy issues
and possible countermeasures.
Index Terms—brain-computer interfaces; BCI applications;
neuromedical; gaming; security; privacy; smartphone
I. INTRODUCTION
Brain-computer interfaces (BCI) are interfaces that collect
data related to users’ brain activities through sensors and
transfer this data to computers. In BCI systems, the brain
does not use peripheral nerves in order to give orders to
our body. Instead, the orders are captured directly by BCI
devices and encoded into electro-physiological signals. These
signals become commands that can control external devices and
computer applications. For example, in order to control a cursor,
signals are transmitted directly from the brain to the application
that moves the cursor, rather than taking the “route” through
peripheral nerves from the brain to the hand to move a mouse.
This technology makes it easier for a human to communicate
with computers or external devices, such as prosthetic devices
(especially for the patients with severe neuromuscular disor-
ders). With the development of intelligentization, BCI technol-
ogy has been pervasive in several fields of our life, such as,
neuromedical field, authentication, gaming, entertainment, and
marketing. Unfortunately, BCI manufacturers are developing
devices and applications without taking much the security and
privacy issues into account. Using such devices, individuals’
private information could be stolen by malicious third party
applications.
Figure 1 shows the working of brain-computer interfaces.
First, the brain neural signals are captured by BCI devices
(step 1): this process is named signal acquisition. After signal
acquisition, BCI systems transform these analog signals into
digital signals (step 2). Then, using signal processing, the
features are extracted and classified (step 3 and step 4). Then,
the signal output is sent to BCI applications (step 5).
Signal
Acquisition
Brain-Computer Interface
Signal Processing
Classification
Feature
Extraction
Preprocessing
(Digitize)
Signal
Output
BCI Applications
1
3
2 4
5
feedback to user
Fig. 1: Working of Brain-Computer Interfaces.
BCI systems could be classified into three main groups: 1)
invasive system [1], 2) partial invasive system, and 3) non-
invasive system. An invasive system requires physical implants
of electrodes into the grey matter of the brain by neurosurgery,
which makes it possible to measure local field potentials. A
partial invasive system (e.g., electrocorticography - ECoG) is
applied to inside of the skull but outside of the grey matter. A
non-invasive system (e.g, electroencephalography - EEG, and
functional Magnetic Resonance Imaging - fMRI) is the most
frequently used neuron signal capturing method. This system
is applied to outside of the skull, just applied on the scalp. It
records the brain activities inside of the skull, and on the surface
of the brain membranes. Both EEG and fMRI give different
perspectives and enable us to “look” inside of the brain [2].
Note that, invasive and partial invasive systems are prone
to scar tissue, and they are difficult to operate. Furthermore,
both of them are quite expensive. Although EEG signals can be
effected by noise and signal distortion, they are easily measured
and have good temporal resolution. Therefore, the most widely
used method for recording brain activity in BCI systems is
EEG. EEG-based devices directly measure electrical potentials
produced by brain’s neural synaptic activities. Five waves from
human brain activities that could be captured by EEG devices
are as follows: 1) gamma waves are in the frequency range of
31Hz and up, and are associated with arousal and excitement
activity of our brain; 2) beta waves are in the frequency range
of 12-30Hz, and are related with action and concentration; 3)
alpha waves are in the frequency range of 8-12Hz, which reflect

relaxation and disengagement; 4) theta waves ranging from 4 to
7Hz, are linked to inefficiency and daydreaming; 5) delta waves
ranging from 0.5 to 4Hz, are the slowest waves and occur when
a user is in hypnoidization.
Currently several companies produce BCI devices, for dif-
ferent purposes, ranging from clinical-grade BCI devices to
consumer-grade BCI devices. Table I lists the main features of
three common devices, while Figure 2 shows the appearances
of these devices.
TABLE I: Comparision of BCI devices.
Device Price Electrodes Resolution Interface
BioSemi Active [3] $12000 256 24 bits Wired
Emotiv EPOC [4] $399-499 14 14 bits Wireless
NeuroSky [5] $50-150 1 8 bits Wireless
(a) Biosemi Active (b) Emotiv EPOC (c) NeuroSky
Fig. 2: BCI devices.
Contribution: In this paper, we discuss the main security
and privacy challenges of brain-computer interfaces with re-
spect to BCI applications. Because of the importance of private
information in our brain (i.e., all of our knowledge, ranging
from passwords to our habits), it is vital to prevent them from
being leaked to attackers. We list the security and privacy
challenges of BCI applications and then discuss their possible
countermeasures.
Organization: The rest of this paper is organized as
follows. In Section II, we revise the main BCI applications
(i.e., neuromedical applications, authentication, gaming and
entertainment, and smartphone-based applications), and for
each application scenario we discuss the key security and
privacy challenges together with possible countermeasures. We
conclude this paper in Section III.
II. BCI SECURITY AND PRIVACY CHALLENGES
In this section, we classify BCI applications into four dif-
ferent application scenarios according to their usage purpose
(i.e., neuromedical applications, authentication, gaming and
entertainment, and smartphone-based applications). For each
application scenario, we provide a description, as well as pos-
sible attacks (either already doable, or envisaged to be possible
in the near future). Finally, for each application scenario we
also discuss possible countermeasures.
A. Neuromedical Applications
Since BCI technology makes it easier for a human to
communicate with external devices or computers, it is widely
used in the neuromedical area to help patients to control their
body through BCI devices instead of nerves. BCI technology
can help patients, especially with serve neurological disorders,
e.g., Parkinson disease. Several neural implantable devices [6]
will be available in the near future. Because of being closely
related with health, security and privacy concerns become
especially necessary to be taken into consideration. An example
of neuromedical applications that might be exposed to attacks
is prosthetic limb application [7], for which, in what follows,
we list possible attacks and countermeasures.
Attacks:As a representative case of neuromedical applica-
tions, prosthetic limb application allows physicians to connect
wirelessly to adjust settings of neural implant devices. If com-
plete brain neural signals are transmitted, an attack can intercept
the transmission, save brain neural signals, and decompose the
raw signals to obtain private information. We underline that
these attacks are possible even when information is transmitted
in an encrypted format [8]. Furthermore, attacker could try
to control prosthetic limbs of patients and give dangerous
movement to patients. Under this condition, an attacker does
not need to be physically in the proximity of the victim. Instead,
the attacker only needs to have attack hardware placed near the
patient. Another possible scenario is the case in which patients
are attackers who might modify settings on their own prosthetic
limbs. They might just want to override mechanical safety
settings to gain extra strength or interfere with limb feedback
to eliminate the ability of feeling pain.
Countermeasures:There are some appropriate safeguards
in the design of the neuromedical applications which can
be deployed in the coming years. For these neuromedical
applications used to give treatment for patients, it is clear
that the main countermeasures should focus on preventing life-
threatening attacks. Also, we should protect private feelings
and emotions of patients from being leaked to attackers. In
addition, these applications should prevent attackers from re-
motely eavesdropping on the wireless signals and collecting
private information about patients’ activities. The communica-
tion between neural implantable devices and patients must be
kept confidential. Furthermore, if they are in sensitive condition
such as depression, trying to prevent wireless attackers from
detecting the presence of these implant devices is effective to
protect safety. In the future, more effective countermeasures
should be proposed to guarantee that neuromedical applications
are not only safe and effective, but also these applications are
robust enough to prevent attacks.
B. User Authentication
Authentication is a process that ensures and confirms a
user’s identity. It plays an important role in security systems.
Using EEG brain signals as authentication measure has been
proposed in many literatures and proved to be effective. Au-
thors in [9] aim at authenticating users, based on brainwave
signals. In particular, they use single-channel EEG signals to
do authentication. In this authentication system, BCI devices
record brainwave signals when a subject performs a custom
task (e.g., singing, breathing or finger movement). Then, brain
signals are wirelessly transmitted to a computer application
which collect and process this data. Their authentication system
analyses the similarity between such brain data and training
data to authenticate subjects. The authors show that their

proposed authentication mechanism has the same accuracy
as multi-channel EEG authentication, about 99% accuracy.
Similar to [9], authors in [10] take EEG brainwave features
as neural passwords to do authentication. The entire process
is performed automatically, without human supervision. The
authors use an algorithm that automatically extracts neural
events corresponding to an individual’s blinking, jaw-clenching,
and eye-rolling activities. The results show that accuracy of this
authentication method ranges from 67% to 95% with single-
trial inputs.
Attacks:Using EEG brainwaves to authenticate might
result in risks for the privacy of users. For example, authors
in [11] propose an authentication system that verifies an indi-
vidual EEG signal when a subject performs a custom task (e.g.,
singing, breathing or finger movement). They also design an
attack model by impersonating the thoughts of subjects. The
authors make deliberate attacks from thought impersonators
to test the robustness of the authentication system. Similar
to [12], an adversary can attack the authentication system via
synthetic EEG signals, using EEG generative model based on
the historical EEG data from a subject can also attack the
authentication system [13].
Countermeasures:To mitigate the authentication attacks
mentioned about, a possible way is to reduce authentication
error rate. For example, we can enlarge the number of partici-
pants, use recruited attackers, and integrate the data processing
methodology with a real-time authentication framework to
achieve reduced authentication error rate. Moreover, another
possible method to enhance the robustness of authentication
is by leveraging multidimensional method [14]. For exam-
ple, using multiple authentication signals (e.g., the signals of
singing, breathing, or being shocked). Besides, we can combine
the existing authentication methods on smartphone device to
perform multidimensional authentication.
C. Gaming and Entertainment
With the development of BCI technology, there are several
BCI games available in entertainment industry [15] [16] [17].
The principle of most BCI games works in a way similar to
P300-speller. In this kind of games, an amplitude peak in the
EEG signal is detected at more or less 300ms after a stimulus.
In the game P300-speller, stimuli are alphanumeric characters
shown on the screen. The characters are arranged in a matrix
where rows and columns flash on a screen in a rapid succession.
According to the being spelled word, users choose one character
using eyes from the screen. Through analyzing peaks occurring
in the brainwaves, authors get the spelled word. Another game
named Snake [18] is also based on the same principle of P300-
speller. In this game, a snake can move in three directions:
forward, left and right. The goal is to locate and eat apples
on a map. Eating apples makes the snake grow in length, and
becomes as large as possible. For the sake of having speed in
the game, moving forward is automatic, and both turning left
or right is controlled by the user via EEG signals.
Attacks:Brain-computer interfaces are becoming increas-
ingly popular in the gaming and entertainment industries. Mar-
tinovic et al. [19] highlight the existence of side-channel attacks
by malicious third-party games on BCI devices. Similar to
smartphone games, third-party BCI games depend on common
APIs to access BCI devices. Thus, such APIs supply unre-
stricted access to raw EEG signals for BCI games. Furthermore,
such games have complete control over the stimuli that can be
presented to users. As a consequence, attackers can display the
contents and read their corresponding EEG signals. The content
might be videos, pictures, or numbers, which users see when
they playing games. Therefore, attackers can specifically design
some videos and images shown to users in order to maximize
the amount of leaked information. In particular, the impact of
exploiting or mishandling BCI devices is difficult to estimate.
Authors in [19] demonstrate BCI games could be exploited to
extract individuals’ private information, such as 4-digit PINs,
bank information, date of birth and location of residence, using
users’ recorded EEG signals.
Countermeasures:Authors in [20] identify security and
privacy issues arising from possible misuse or inappropriate use
of “Brain Malware” information. In particular, they propose
an interdisciplinary approach to enhance the security of BCI
systems by the aid of several experts from different areas, such
as neuroscientists, neural engineers, ethicists, as well as legal,
security and privacy experts.
Authors in [21] propose a tool named “BCI Anonymizer”
to prevent the side-channel extraction of users’ private in-
formation. The basic idea of the “BCI Anonymizer” is to
remove private information from raw EEG signals before this
information is stored and transmitted. “BCI Anonymizer” could
be implemented either in hardware or in software, as a part
of BCI devices, but not as part of any external network or
computational platform. Moreover, the “BCI Anonymizer” can
generate an anonymized neural signals to replace the removed
signals that represent private information. However, authors
in [21] do not provide a clear method to distinguish the
difference between users’ private information and commands
to applications.
D. Smartphone-based Applications
The application scenario we want to consider in this section
is actually mainly driven by a specific emerging and pervasive
technology, i.e., smartphones. Along with the advances in
smartphone capabilities, there is an increasing interest in using
smartphone by individuals in their daily life. BCI are used in
conjunction with this technology (smartphone). Recently, some
BCI applications based on smartphone have been proposed in
many literatures.
Authors in [22] implement a brain-controlled address book
dialing app, which works in a way similar to P300-speller.
Instead of showing characters in P300-speller, the dialing app
shows a sequence of photos of contacts from the address book.
Therefore, the user can easily select a person whom she or he
wishes to dial. Authors in [23] measure a subject’s attention
and meditation level through EEG signals when a subject is
playing a game. Authors compare the difference among all
the subjects’ EEG signals, according to subjects’ age and
gender. Their results show that, in the POKOPANG game, the
average attention level of men is lower than that of women,
while the meditation level is reversed. As a result, authors
infer that women are more interested in POKOPANG game.

Air Brain system [24] is a portable EEG telemetry system.
Different from other portable EEG monitoring systems, in order
to have more storage space, this way, the stored data can be
accessed from everywhere. To achieve this, the system uses
3G network of smartphone to transfer data. Air Brain system
enables subjects to measure EEG signals immediately after
subjects start walking. Furthermore, the system is able to detect
eye closing by measuring changes of aplha wave.
Attacks:The smartphone-based BCI applications are
prone to attacks that originate in the mobile device itself.
Therefore, most of the possible attacks on smartphone issues
could also be considered as security and privacy issues of
smartphone-based BCI applications. These applications can
access private data which is acquired from BCI devices and
stored in smartphones or SD card. This data can be illegally
transferred by a malware to a remote server (e.g., privilege
escalation attacks [25]). Developers of malwares can analyse
the private signals and attack the users of BCI devices. Attacks
to smartphone applications also apply to smartphone-based BCI
applications.
Countermeasure:Given that we are considering BCI
applications in conjunction with a specific technology (smart-
phone), here countermeasures are mostly the ones typical for
generic smartphone security. Useful security approaches could
be the ones that track the flow of information. For example,
TaintDroid [26] proposes a model that can track not only the
way applications access sensitive data, but also the way appli-
cations use such data. FlowDroid [27] proposes an innovative
and accurate static taint analysis for applications in Android
platform, allowing proper analysis to handle callbacks invoked
by the Android framework. In addition to the aforementioned
approaches, fine-grained context-based access control [28] is
another effective way to limit the leakage of private data.
These mitigations are possible only by modifying Android’s
permission model, e.g., Android’s internal middleware layer.
III. CONCLUSION
In this paper, we survey some common brain-computer
interfaces (BCI) applications, and their possible security and
privacy issues. Moreover, we consider four different application
scenarios: 1) neuromedical applications, 2) user authentication,
3) gaming and entertainment, and 4) smartphone-based appli-
cations. For each scenario we provide the description of current
state-of-the-art technologies, potential attacks that might threat
each scenario, and envisaged countermeasures.
ACKNOWLEDGMENTS
Mauro Conti is supported by a European Marie Curie Fellow-
ship (N. PCIG11-GA-2012-321980). This work is also partially
supported by the Italian MIUR PRIN Project TENACE (N.
20103P34XC), and the University of Padua PRAT 2014 Project
on Mobile Malware.
REFERENCES
[1] J. R. Wolpaw, N. Birbaumer, W. J. Heetderks et al., “Brain-computer
interface technology: a review of the first international meeting,” IEEE
transactions on rehabilitation engineering, vol. 8, no. 2, pp. 164–173,
2000.
[2] J. Kropotov, Quantitative EEG, event-related potentials and neurotherapy.
Academic Press, 2010.
[3] (2015, July) Biosemi. [Online]. Available: http://www.biosemi.com
[4] (2015, July) Emotiv epoc. [Online]. Available: https://emotiv.com
[5] (2015, July) Neurosky. [Online]. Available: http://neurosky.com
[6] T. Denning, Y. Matsuoka, and T. Kohno, “Neurosecurity: security and
privacy for neural devices,” Neurosurgical Focus, vol. 27, no. 1, p. E7,
2009.
[7] A. B. Schwartz, X. T. Cui, D. Weber, and D. W. Moran, “Brain-
controlled interfaces: Movement restoration with neural prosthetics,”
Neuron, vol. 52, no. 1, pp. 205 – 220, 2006.
[8] M. Conti, L. V. Mancini, R. Spolaor, and N. V. Verde, “Can’t you hear
me knocking: Identification of user actions on android apps via traffic
analysis,” in DASP, 2015, pp. 297–304.
[9] J. Chuang, H. Nguyen, C. Wang, and B. Johnson, “I think, therefore i am:
Usability and security of authentication using brainwaves,” in Financial
Cryptography and Data Security, 2013, pp. 1–16.
[10] A. Rajagopal, A. C. Nguyen, and D. M. Briggs, “Neuropass: A secure
neural password based on EEG,” in Biomedical Engineering, 2013.
[11] B. Johnson, T. Maillart, and J. Chuang, “My thoughts are not your
thoughts,” in Proceedings of the 2014 ACM UbiComp: Adjunct Publi-
cation, 2014, pp. 1329–1338.
[12] P. E. McSharry, G. D. Clifford, L. Tarassenko et al., “A dynamical
model for generating synthetic electrocardiogram signals,” Biomedical
Engineering, vol. 50, no. 3, pp. 289–294, 2003.
[13] S. T. Archer and B. D. Pless, “Stimulation signal generator for an
implantable device,” Feb. 10 2004, uS Patent 6,690,974.
[14] T. Naik and S. Koul, “Multi-dimensional and multi-level authentication
techniques,” International Journal of Computer Applications, vol. 75,
no. 12, pp. 17–22, 2013.
[15] C. M¨
uhl, H. G¨
urk¨
ok, D. Plass-Oude Bos, M. E. Thurlings et al., “Bacteria
hunt: A multimodal, multiparadigm bci game,” University of Genua, 2010.
[16] M. Congedo, M. Goyat, N. Tarrin, and G. e. a. Ionescu, “Brain invaders:
a prototype of an open-source p300-based video game working with the
openvibe platform,” in 5th International BCI, 2011, pp. 280–283.
[17] A. Finke, A. Lenhardt, and H. Ritter, “The mindgame: a p300-based
brain–computer interface game,” Neural Networks, vol. 22, no. 9, pp.
1329–1333, 2009.
[18] E. A. Larsen, “Classification of eeg signals in a brain-computer interface
system.” Norwegian University, 2011.
[19] I. Martinovic, D. Davies, M. Frank, D. Perito, T. Ros, and D. Song, “On
the feasibility of side-channel attacks with brain-computer interfaces,” in
USENIX Security 12, 2012, pp. 143–158.
[20] T. Bonaci, R. Calo, and H. J. Chizeck, “App stores for the brain: Privacy
& security in brain-computer interfaces,” in Science, Technology and
Engineering, 2014 IEEE International Symposium, 2014, pp. 1–7.
[21] H. Chizeck and T. Bonaci, “Brain-computer interface anonymizer,”
Aug. 14 2014, US Patent App. 14/174,818. [Online]. Available:
http://www.google.com/patents/US20140228701
[22] A. Campbell, T. Choudhury, S. Hu, H. Lu et al., “Neurophone: brain-
mobile phone interface using a wireless eeg headset,” in Proceedings of
the second ACM SIGCOMM workshop, 2010, pp. 3–8.
[23] J.-Y. Kim and W.-H. Lee, “Eeg signal feature analysis of smartphone
game user,” ASTL, vol. 39, pp. 14–19, 2013.
[24] K. Honda and S. N. Kudoh, “Air brain: the easy telemetric system with
smartphone for eeg signal and human behavior,” in Proceedings of the
8th BodyNets, 2013, pp. 343–346.
[25] L. Davi, A. Dmitrienko, A.-R. Sadeghi, and M. Winandy, “Privilege
escalation attacks on Android,” pp. 346–360, 2011.
[26] W. Enck, P. Gilbert, S. Han, and V. e. a. Tendulkar, “Taintdroid: an
information-flow tracking system for realtime privacy monitoring on
smartphones,” ACM TOCS, vol. 32, no. 2, p. 5, 2014.
[27] S. Arzt, S. Rasthofer, C. Fritz, E. Bodden, and et al., “Flowdroid: Precise
context, flow, field, object-sensitive and lifecycle-aware taint analysis for
android apps,” in ACM SIGPLAN Notices, 2014, pp. 259–269.
[28] M. Conti, B. Crispo, E. Fernandes, and Y. Zhauniarovich, “Crˆ
epe: A
system for enforcing fine-grained context-related policies on android,”
Information Forensics and Security, vol. 7, no. 5, pp. 1426–1438, 2012.