Page 104 - Artificial Intelligence for the Internet of Everything

90 Artificial Intelligence for the Internet of Everything

To-date, the cybersecurity engineering community has principally been
focused on information systems, an area where the risks are different and
the technical factors regarding cyber defense pose significantly different
challenges.

5.2.1 Historic Patterns for Addressing Cybersecurity

While cybersecurity experts point to the fact that incorporating anticipatory
cyber-security features into the design of systems provides a pathway for
achieving better security, historically most solutions have been add-ons to
systems in response to actual attacks (Miller, 2014b). The reasons for this
are economic. When new innovations are in their early development phase
(such as autonomous vehicles), designers are consumed with achieving a
working system, and security is treated as something that will follow. When
the innovation is ready to be brought to the market, concern about the cost
impacts the security of the new products’ prices and further delays security
implementation. When the new products are selling, but significant attacks
have yet to occur, there is no pressing demand to anticipate attacks. When
attacks start occurring, and there are already large numbers of existing sys-
tems in use, responsive patching becomes the de-facto solution.
For existing information systems the major consequences of cyber attacks
have been financial in nature or related to privacy. Should human safety
become a primary risk of cyber attacks in the future, new societal patterns
may emerge that demand stronger anticipatory solutions. Anticipatory solu-
tions must be designed not only on the basis of prior attacks, but also based
upon predictions of what cyber attackers might target in the future and how
they might implement these attacks. Prediction of attacker behavior is quite
complex, requiring considerations such as: (1) historic attacks; (2) attacker
motivations; (3) attack complexity and corresponding attacker skill require-
ments; (4) costs of design and implementation; (5) risks of attacks failing; and
(6) risks of getting caught. This situation is exacerbated by the need for com-
petitors to share information (e.g., historic attack information) in order to
have a more complete basis for making predictions and to provide the
opportunity to derive a common framework for considering solutions that
are related to a domain of similar products. Furthermore, for physical systems
classes that include rapidly changing automation features, predictions can be
unstable (e.g., the increasing rate for adding new automation features in
automobiles points to the need for annual reconsideration of potential
cyber attacks and the corresponding defenses). This situation is further