IEEE Consumer Electronics Magazine - January/February 2024 - 66

Special Section on Security and Privacy-Aware Emerging Computing
FIGURE 2. Deepfakes-Authentic content to potential intended manipulation(s) .
literature. The term " supervised " here refers to
the fact that the system facilitating the identity
establishment (e.g., kiosk used for streaming
video to a remote operator) is monitored by the
service provider, and is difficult to alter/modify
by an applicant or potential adversary to circumvent
the identity proofing mechanism.
The physical presence requirement for conventional
identity proofing is thus replaced by
requiring applicants to visit a dedicated facility to
speak with another person remotely. The other
requirements for supervised remote in-person
identity proofing at IAL3, as per NIST,2 include
1) Live participation by a (remote) operator
throughout the entire identify proofing process,
without allowing the applicant to
depart from this process at any stage.
2) All actions taken by the applicant are clearly
visible to the remotely connected operator.
3) The operator is appropriately trained to
detect any potential instances of fraud and
to properly conduct the supervised remote
in-person identity proofing.
4) Digital identity verification (e.g., through a
chip or any wireless technology) is possible
through the embedded sensors and scanners
(at the supervised facility).
5) Physical tamper detection and resistance features
are in place and are appropriate for the
deployed environment.
6) All communication occurring over a mutually
authenticated channel.
While IAL2 can be achieved under
" unsupervised remote " conditions, meaning that
neither a trained operator nor travel to a specific
facility is required, it is considered to offer a
lower level of assurance than IAL3 due to the lack
of human oversight. The premise is that
machines cannot yet match human intelligence,
although AI is quickly approaching this standard.
The Advancement of Deepfake Methods
As highlighted by ENISA,1 deepfake imagery
generated by advanced video manipulation technologies
already poses a serious threat to the
trustworthiness of remote identity proofing. The
term " deepfake " combines the concept of " deep
learning, " made possible through advanced artificial
neural networks (ANNs), with the word
" fake " to describe fabricated media content generated
through a carefully crafted " deep neural
network " (often referred to as AI). As shown in
Figure 2, deepfakes can be created for either
innocuous entertainment or malicious purposes.
The latter include tampering with evidence and
spoofing, both of which are relevant to remote
identity proofing-particularly in cases where
an adversary is in control of the device from
which media is being streamed for identity
establishment. It should be noted that deepfakes
are distinct from adversarial machine learning,
which aims to deceive machines only.
By contrast, deepfakes are often challenging
to identify for both humans and machines, making
them a conceivable tool for subverting the
66
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2024

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2024
