Content uploaded by Shadrack Awah Buo
Author content
All content in this area was uploaded by Shadrack Awah Buo on Dec 14, 2020
Content may be subject to copyright.
The Emerging Threats of Deepfake Attacks and
Countermeasures
Shadrack Awah Buo
Department of Computing & Informatics
Bournemouth University, UK
Abstract—Deepfake technology (DT) has taken a new
level of sophistication. Cybercriminals now can
manipulate sounds, images, and videos to defraud and
misinform individuals and businesses. This represents a
growing threat to international institutions and
individuals which needs to be addressed. This paper
provides an overview of deepfakes, their benefits to
society, and how DT works. Highlights the threats that
are presented by deepfakes to businesses, politics, and
judicial systems worldwide. Additionally, the paper will
explore potential solutions to deepfakes and conclude
with future research direction.
Keywords—Deepfakes, Artificial Intelligence, cybercrime,
multi-media manipulation.
1. INTRODUCTION
Developments in Artificial Intelligence (AI) have led to the
emergence of deepfake technologies (DT), which pose a
significant threat to global institutions. Deepfake is defined
as an AI-based technology that can manipulate images,
sounds, and video content to represent an event that did not
occur [1]. For instance, the faces of politicians being edited
onto other individuals’ bodies who appear to say things that
they never did are becoming commonplace. This growing
phenomenon has been used in political scenarios to
misinform the public on various debates [2]. For instance,
the use of deepfake video by an Italian satirical TV show
against the formal Prime Minister of Italy Matteo Renzi.
The video shared on social media depicted him insulting
fellow politicians. As the video spread online, many
individuals started to believe the video was authentic, which
led to public outrage [2]. Additionally, cybercriminals have
used deepfakes to impersonate Chief Executive Officers
(CEOs) in companies to deceive employees, usually from
finance departments to transfer funds to bank accounts
controlled by the scammers [3]. Most deepfake
manipulations are intended for entertainment purposes such
as movies, videogames, and educational videos. However,
cybercriminals have taken advantage of the technology to
misinform and defraud businesses and the individual [4].
Moreover, creating such deepfakes media requires expertise
and specialist software and hardware. However, freely
available tools [22] such as “FaceSwap” and “Reface” have
enabled unskilled individuals to participate in media
manipulation for entertainment or malicious purposes.
There are some questions that this paper will aim to
investigate. Firstly, what are the negative implications of DT
for global institutions? What are the impacts of DT on these
institutions? Lastly, how can they mitigate the threats of
DT?
This paper examines the threats posed by deepfakes that
affect global institutions and the negative implications
(section 4). It also provides an overview of deepfakes
(section 2), how they are created and the benefits of its use
(section 3). The potential solutions to prevent deepfakes and
what can be done to mitigate future risks are also discussed
(section 5).
2. AN OVERVIEW OF DEEPFAKES
The term deepfake is a combination of “deep learning” and
“fake” [5]. The deepfake phenomenon started on the social
media platform, “Reddit.” An anonymous user shared an
altered pornographic video of a celebrity, their face had been
swapped with a porn actor. Even though the user was banned
from Reddit, Kirchengast [7] argues that their actions
sparked an increased interest in deepfakes, and new content
began to spread on other social media platforms like
“Twitter” and “4chan”. Since the inception of DT, it has
been utilized by hobbyists to manipulate multi-media content
by matching human expressions and tones to create media
that appears authentic [9]. A famous example was created by
the comedian Jordan Peele which depicted the former US
President Barack Obama delivering a speech to raise
awareness about the threats of DT [9].
DT is powered by Generative Adversarial Networks (GANs).
GANs employs two Artificial Neural Networks (ANNs)
working together to create deepfakes. These ANNs are also
known as “detector” or discriminative network, and
“synthesizer” or generative network [5]. They are trained on
a large dataset of videos, images, and sounds to produce
high-quality deepfakes [5]. The synthesizer initiates the
sequence by generating deepfake content that is accurate
enough to trick the detector. The detector is responsible for
analysing and distinguishing whether the deepfake produced
by the generator appears authentic. The cycle is continued
until the discriminator is unable to detect media forgery,
thereby improving the overall quality of the deepfake before
it can be deployed [4]. It is expected that future GANs
algorithms will be trained on smaller datasets and produce
more convincing, higher quality deepfakes [11]. Therefore,
these developments would allow cybercriminals to create
more authentic deepfakes which would have a devastating
impact on their victims. Aside from the risks that DT can
pose, the next section looks at some of the benefits and
positive applications that DT can provide for society.
3. BENEFITS OF DEEPFAKE TECHNOLOGY
Despite the malicious usage of DT, there are positive
applications. For instance, voice assistant technologies such
as Apple’s Siri and Window’s Cortana uses machine learning
(ML). These technologies apply similar AI-based algorithms
to assist the end-user by answering queries and delivering
content by voice-activated commands [7]. Moreover, Google
has developed an AI tool called “TensorFlow” to facilitate
the discovery of content that is relevant to search requests in
Gmail and Google Translate [10].
According to Chesney and Citron [12], the education sector
could also benefit from DT, by presenting students with
information in compelling ways. For example, being used to
recreate historical figures and events to improve student
participation in history lessons [12] [32]. Ongoing research
[13] is exploring ways to develop an AI system that will
automate the process of producing educational content using
DT. One system in particular is known as “LumièreNet” will
streamline the process of creating educational videos and
presentations on learning platforms such as Udacity [13].
Moreover, DT can be used in the gaming industry to improve
the player’s experience. For example, it can be used to
develop realistic virtual environments and natural-sounding
in-game assistants which improve the user experience [4].
Additionally, DT has enabled film producers to cast
deceased actors such as Paul Walker, who died before the
completion of the movie “Furious 7”. DT was used to
recreate his face for the last scene of the movie [32].
Similarly, DT can have positive uses in health and social
care. For instance, it can help individuals deal with the loss
of loved ones by developing a digital version of their loved
ones [4].
DT can also assist in the rehabilitation process with
individuals who suffer from addictive habits, such as
smoking. The World Health Organisation [14] has developed
an AI-based solution called “Florence” which helps
individuals cope with tobacco addiction. Users can have a
virtual conversation with “Florence” to build confidence to
quit smoking by creating a plan to track their process.
Researchers are also exploring the use of GANs to detect
abnormalities in X-rays and their potential to detect early
signs of diseases [15]. This will assist medical professionals
in the early detection of illnesses and potentially save more
lives.
4. THREATS OF DEEPFAKES TECHNOLOGY
In contrast to the benefits of DT pose major threats to global
institutions for the following reasons; they used to defraud
businesses and raise cybersecurity concerns to organisations.
They can also, be used as a source of misinformation in
politics and the court of law. This section explores the threats
of deepfakes to Businesses, politics, and judicial systems.
4.1. Threats to Judicial Systems
According to Pfefferkorn [22], evidence tampering is one of
the major threats posed by deepfakes in the judicial system.
Evidence in the court of law can be manipulated with the use
of DT to sway a case one way or the other. Further issues
may arise during cross-examinations when an offering party
testifies affirmatively concerning details of a deepfake video
while the opposing party denies the contents of the video
[22]. This would negatively impact court cases because
deepfakes might cause additional caseloads, and cost money
and time to verify and authenticate the evidence before it can
be admissible in court [22]. For example, in a UK child
custody case, a deepfake audio file was presented as
evidence to the court by the mother [8]. The mother had used
DT and tutorials online to create a plausible audio file that
sounded like a recording of the father threatening her, to
support her claim that he was too violent to be allowed
access to their children. However, after the file was
forensically examined, it was proven to be fake and
dismissed by the courts.
Currently, legal systems are not equipped to effectively
combat deepfakes used in evidence tampering [23]. The
UK’s Civil Evidence Act 1968 [24], states that video
recordings are admissible in court if the video is deemed
trustworthy. However, it is becoming increasingly difficult to
distinguish deepfakes from the original content. In some
instances, deepfake detection measures have failed to detect
the use of deepfakes [16]. Therefore, there is an urgent need
for new and effective countermeasures to prevent evidence
tampering in future court cases.
4.2. Threats to politics
Another threat that can strengthen with the use of deepfakes
is disinformation within politics [17]. Deepfakes can be
quickly created and easily circulated to a wide audience.
With this specific advantage, DT can be used knowingly or
unknowingly to misinform the public for political advantage
[17]. One notable example was the circulation of an altered
video of an American politician, Nancy Pelosi on social
media. In the video, she appeared intoxicated while
mispronouncing her words [18]. The American President
Donald J. Trump shared the video on his “Twitter” account
to alter public perception of his opponent, Nancy Pelosi.
Consequently, the video had been viewed and shared over
2.5 million times on Facebook [19]. Despite bipartisan calls
for the video to be taken down, a Facebook spokesperson
confirmed that the videos will not be removed because the
platform does not have policies that dictate the removal of
false information [20]. Therefore, this has prompted world
governments to look for ways to regulate the use of DT [7].
Additionally, deepfakes can have a damaging impact on
geopolitics and relationships between countries. Recently,
the Australian Prime Minister, Scott Morrison demanded an
apology after Zhao Lijian, a spokesperson for China’s
foreign ministry posted a fake image on “Twitter” that
depicted an Australian soldier holding a knife to the throat of
an Afghan child [25]. The image sparked outrage online and
a bitter debate between the Chinese and Australian
governments. Moreover, this incident is likely to worsen
diplomatic relationships between the two countries [25].
Therefore, there is a growing need for policies to regulate the
use of deepfakes on social media platforms for political gain.
4.3. Threats to Businesses
In addition to its impact on the legal system and politics, DT
can have an adverse budgetary impact on businesses.
Combined with social engineering attacks such as email
phishing, deepfakes can be used to defraud businesses which
can affect inter-business negotiations and the organisation’s
reputation [9]. For instance, scammers can impersonate
senior figures to obtain sensitive information or request
money transfers without detection. In one instance,
scammers defrauded a UK (United Kingdom) based firm by
impersonating the Chief Executive Officer (CEO) [26]. He
convinced employees from the finance department to transfer
$220,000 to an account controlled by the scammers [26].
Symantec, a cybersecurity company, revealed that deepfakes
and social engineering was used to defraud three CFOs
(Chief Financial Officer) of undisclosed substantial funds
[29]. In addition to these findings, Forrester Research [29]
predicted a monetary loss of $250 million by the end of 2020
from deepfake frauds. With the continuous advancement of
DT, businesses are likely to continue suffering considerable
financial losses from deepfake scams.
Deepfakes could also negatively impact organisations that
incorporate the use of Biometric technology [1]. Companies
have started to adopt biometric technology as a security
measure in the workplace [1]. For example, the installation
of face scanners to grant access to restricted areas. However,
if these areas are breached with the use of deepfakes, this
could lead to unauthorised access to sensitive information
and intellectual property. Such an attack could lead to
monetary loss due to the costs incurred from containing the
breach, compensating customers, and heighten security costs
[30].
5. POTENTIAL SOLUTIONS TO DEEPFAKES
Recently, a multitude of solutions have been proposed and
deployed against deepfakes. A recent study [27] revealed that
current DT mostly generates low-resolution media contents,
which are easy to identify using Convolutional Neural
Networks (CNNs). Using CNNs, researchers [27] were able
to successfully detect and identify 99.1% of deepfakes.
Despite positive results from these experiments, the research
conceded that CNN's should not be completely relied upon,
since the failure rates will factor in during some instances.
Additionally, with the growing number of high-quality
deepfakes, current CNNs will become ineffective [28].
According to Albahar and Almalki [31], digital forensics can
provide an effective solution for detecting deepfakes. Using
computational techniques, forensics experts can observe
whether image pixels have been altered, by isolating
anomalies, such as shadows and reflections [33]. They can
also inspect the metadata of the file to check if it has been
altered, by checking the edit history and how many times the
file has been compressed. However, having access to a
dedicated forensics team and expert tools needed for
detecting deepfakes can be costly to manage and maintain
[34]. To solve this issue, Lee and Un [34] proposed digital
forensics as a service model, which leverages cloud
computing technologies to provide robust forensics services
at a cheaper price.
Furthermore, Hasan and Salah [28], proposed a solution
based on traceability and transparency rather than detection.
Using blockchain and smart contracts, Hasan, and Salah’s
solution acts as a transparent digital signature on media
content to prove their authenticity. This solution relies on
time-sequence logs to track the history of media contents,
monitoring where it was used online to later determine their
origins [28]. This solution can be easily integrated into a web
browser to indicate the authenticity of multi-media contents
online. Admittedly, the solution has drawbacks which might
negate its effectiveness. Muna [37] argues that this solution
would be prone to errors, such as falsely identifying media
content as fake. Additionally, Blockchain and smart contracts
are relatively new technology, and the concept might be
expensive and difficult to implement.
Despite these technological breakthroughs, it is not a full-
proof system since there is a chance some deepfakes cannot
be detected. Moreover, Lyu and Li [27] admitted that more
research and development is required in detection technology
since deepfakes will continue to evolve. Therefore, it is
essential to adopt solutions geared towards preventing the
issues from occurring, by advocating for employee education
and awareness training.
Westerlund [4] suggest that employees could be trained to
identify whether the information being displayed is
legitimate or falsified. For instance, businesses can establish
a two-step authentication policy which encourages
employees to verify requests from phone calls and emails or
have a second employee verifying funds transfers [29].
Businesses can further strengthen their security measures by
limiting data accessibility to images and videos on social
media platforms. This would prevent cybercriminals from
utilizing such data to create deepfakes [29].
Another solution was put forward by Meskys [35] who
suggests that technology firms and governments should
consider imposing sanctions and regulations on the creation
of socially harmful deepfakes. This will prevent the spread of
misinformation and defamation of character [6][21].
However, Hall [36] contends that regulating deepfakes will
have negative implications on freedom of expression, adding
that establishing legal rules will push too far into censorship.
Hence it is crucial to avoid blanket implementations of
regulations that could infringe on freedom of expression.
6. CONCLUSION
In conclusion, the continuous evolution of cybercrime has
culminated with deepfakes which severely magnify the
threats of traditional frauds. DT continues to pose various
threats such as misinformation in politics, fraud, and
evidence tampering in court. Existing technical solutions can
be implemented to prevent deepfakes attacks; however, these
methods should not be solely relied upon to tackle the issue
of deepfakes. Therefore, it is also critical to invest in
awareness and training to help identify early signs of
deepfake attacks. Technology firms and governments should
consider passing legislation that will criminalize the use of
deepfakes with the intent to defame the character of
individuals. This ensures that appropriate punishments and
consequences are taken for malicious users. In terms of
future research, Facebook [38] has partnered with Microsoft
and issued a public challenge worth $10 million, that will
help produce a technology that can be us by everyone to
detect deepfakes. This process of crowdsourcing knowledge
will help techs companies develop an effective solution
against deepfake attacks.
7. REFERENCES
[1]
J. Wojewidka, “The deepfake threat to face biometrics,” Biometric Technology Today,
vol. 2020, no. 2, pp. 5-7, 2020.
[2]
S. Venkataramakrishnan, “Can you believe your eyes? How deepfakes are coming for
politics,” 2019. [Online]. Available: https://www.ft.com/content/4bf4277c-f527-11e9-
a79c-bc9acae3b654. [Accessed 20 November 2020].
[3]
C. Stupp, “Fraudsters Used AI to Mimic CEO’s Voice in Unusual Cybercrime Case,” 30
October 2019. [Online]. Available: http://fully-human.org/wp-
content/uploads/2019/10/Stupp_Fraudsters-Used-AI-to-Mimic-CEOs-Voice-in-Unusual-
Cybercrime-Case.pdf. [Accessed 20 November 2020].
[4]
M. Westerlund, “The Emergence of Deepfake Technology: A Review,” Technology
Innovation Management Review, vol. 9, no. 11, pp. 40-53, 2019.
[5]
Rana, S; Sung, AH; “DeepfakeStack: A Deep Ensemble-based Learning Technique for
Deepfake Detection,” New York, 2020.
[6]
J. Kietzmann, L. W. Lee, I. P. McCarthy, and T. C. Kietzmann, “Deepfakes: Trick or
treat?,” Business Horizons, vol. 63, no. 2, pp. 135-146, 2020.
[7]
T. Kirchengast, “Deepfakes and image manipulation: criminalisation and control.,”
Information & Communications Technology Law, vol. 29, no. 3, pp. 308-323, 2020.
[8]
Daily Telegraph, “Doctored audio evidence used to damn father in custody battle.,” 2020.
[Online]. Available:
https://link.gale.com/apps/doc/A612903662/STND?u=bu_uk&sid=STND&xid=c86cb1bc
. [Accessed 27 November 2020].
[9]
C. Owen-Jackson, “What does the rise of deepfakes mean for the future of
cybersecurity?,” 2020. [Online]. Available: https://www.kaspersky.com/blog/secure-
futures-magazine/deepfakes-2019/28954/. [Accessed 20 November 2020].
[10]
Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., Devin, M., Ghemawat, S.,
Irving, G., Isard, M. and Kudlur, M., “TensorFlow: A System for Large-Scale Machine
Learning,” USENIX, Savannah, 2016.
[11]
Pan, Z; Yu, W; Yi, X; Khan, A; Yuan, F; Zheng, Y; “Recent Progress on Generative
Adversarial Networks (GANs): A Survey,” IEEE Access, vol. 7, pp. 36322 - 36333, 2019.
[12]
Chesney, B; Citron, D; “Deep fakes: A looming challenge for privacy, democracy, and
national security,” California Law Review, vol. 107, no. 6, pp. 1769-1771, 2019.
[13]
B. Kim and V. Ganapathi, “LumièreNet: Lecture Video Synthesis from Audio,” 04 July
2019. [Online]. Available: https://arxiv.org/abs/1907.02253. [Accessed 20 November
2020].
[14]
World Health Organisation (WHO), “Quit tobacco today!,” 2020. [Online]. Available:
https://www.who.int/news-room/spotlight/using-ai-to-quit-tobacco. [Accessed 21
November 2020].
[15]
Kazeminia, S; Baur, C; Kuijper, A; Ginneken, B V; Navab, N; Albarqouni, S;
Mukhopadhyay, A;, “GANs for medical image analysis,” Elsevier, vol. 109, no. 2020,
2020.
[16]
P. Korshunov and S. Marcel, “Vulnerability assessment and detection of Deepfake
videos,” Crete, 2019.
[17]
Gosse, C; Burkell, J;, “Politics and porn: how news media characterizes problems
presented by deepfakes,” Critical Studies in Media Communication, vol. 37, no. 5, pp.
497-511, 2020.
[18]
Reuters , “Trump retweets doctored video of Pelosi to portray her as having 'lost it',”
2019. [Online]. Available: https://www.reuters.com/article/us-usa-trump-pelosi-
idUSKCN1SU2CB. [Accessed 29 November 2020].
[19]
S. Greengard, “Will deepfakes do deep damage?,” Communications of the ACM, vol. 63,
no. 1, pp. 17-19, 2019.
[20]
M. Kelly, “Distorted Nancy Pelosi videos show platforms aren’t ready to fight dirty
campaign tricks,” 2019. [Online]. Available:
https://www.theverge.com/2019/5/24/18637771/nancy-pelosi-congress-deepfake-video-
facebook-twitter-youtube. [Accessed 29 November 2020].
[21]
A. Yadlin-Segal and Y. Oppenheim, “Whose dystopia is it anyway? Deepfakes and social
media regulation,” Convergence, pp. 3-14, 2020.
[22]
R. Pfefferkorn, “"deepfakes" in the courtroom,” Boston University Law Journal, vol. 29,
no. 2, pp. 245-276, 2020.
[23]
A. Alexandrou and M. H. Maras, “Determining authenticity of video evidence in the age
of artificial intelligence and in the wake of Deepfake videos,” The International Journal
of Evidence & Proof, vol. 23, no. 3, pp. 255-262, 2018.
[24]
Legislation.gov.uk, “Civil Evidence Act 1968,” 1968. [Online]. Available:
https://www.legislation.gov.uk/ukpga/1968/64/pdfs/ukpga_19680064_en.pdf. [Accessed
11 November 2020].
[25]
BBC, “Australia demands China apologise for posting 'repugnant' fake image,” 2020.
[Online]. Available: https://www.bbc.co.uk/news/world-australia-55126569. [Accessed
01 December 2020].
[26]
The Wall Street Journal, “Fraudsters Used AI to Mimic CEO’s Voice in Unusual
Cybercrime Case,” 2019. [Online]. Available: https://www.wsj.com/articles/fraudsters-
use-ai-to-mimic-ceos-voice-in-unusual-cybercrime-case-11567157402. [Accessed 10
November 2020].
[27]
Li, Y; Lyu, S; “Exposing DeepFake Videos By Detecting Face Warping Artifacts,” 22
May 2019. [Online]. Available: https://arxiv.org/abs/1811.00656. [Accessed 22
November 2020].
[28]
Hasan, H R; Salah, K; “Combating Deepfake Videos Using Blockchain and smart
contracts,” IEEE Access, vol. 7, pp. 41596 - 41606, 2019.
[29]
S. Sjouwerman, “The evolution of deepfakes: Fighting the next big threat,” 2019.
[Online]. Available: https://techbeacon.com/security/evolution-deepfakes-fighting-next-
big-threat. [Accessed 09 November 2020].
[30]
Rosati, P; Cummins, M; Deeney, P; Gogolin, F; der Werff, L v; Lynn, T; “The effect of
data breach announcements beyond the stock price: Empirical evidence on market
activity,” International Review of Financial Analysis, vol. 49, no. 2017, pp. 147-148,
2017.
[31]
Albahar, M. and Almalki, J., “Deepfakes: Threats and countermeasures systematic
review,” Journal of Theoretical and Applied Information Technology, vol. 97, no. 22, pp.
3242-3250, 2019.
[32]
S. Salmani and D. Yadav, “Deepfake: A Survey on Facial Forgery Technique Using
Generative Adversarial Network,” in 2019 International Conference on Intelligent
Computing and Control Systems (ICCS), Madurai, 2020.
[33]
K. Hao, “Deepfake-busting apps can spot even a single pixel out of place,” MIT
Technology Review, 01 November 2018. [Online]. Available:
https://www.technologyreview.com/2018/11/01/139227/deepfake-busting-apps-can-spot-
even-a-single-pixel-out-of-place/. [Accessed 29 November 2020].
[34]
J. Lee and S. Un, “Digital forensics as a service: A case study of forensic indexed
search,” in 2012 International Conference on ICT Convergence (ICTC), Jeju Island,
2012.
[35]
E. Meskys, A. Liaudanskas, J. Kalpokiene and P. Jurcys, “Regulating deep fakes: legal
and ethical considerations,” Journal of Intellectual Property Law & Practice, vol. 15, no.
1, pp. 24-31, 2020.
[36]
H. K. Hall, “DEEPFAKE VIDEOS: WHEN SEEING ISN'T BELIEVING,” 'Deepfake
Videos: When Seeing Isn't Believing' Catholic University Journal of Law and
Technology., vol. 27, no. 1, pp. 51-76, 2018.
[37]
M. Muna, "Technological Arming: Is Deepfake the Next Digital Weapon?,"
ResearchGate, Berkeley, 2020
[38]
Facebook, "Creating a data set and a challenge for deepfakes," Facebook AI, 05
September 2019. [Online]. Available: https://bit.ly/3npque0. [Accessed 01 December
2020].
