Three Ethical Challenges of Applications of Artificial Intelligence in Cybersecurity

In 2017, the WannCry and NotPeya showed that attacks targeting the cyber component of infrastructures (e.g. attacks on power plants), services (e.g. attacks to banks or hospitals servers), and endpoint devices (e.g. attacks on mobiles and personal computers) have a great disruptive potential and could cause serious damage to our information societies. WannaCry crippled hundreds of IT systems. And NotPetya costed pharmaceutical giant Merck, shipping firm Maersk and logistics company FedEx around US$300 million each. At a global level, cyber crime causes multibillion dollar losses to businesses, with average losses per organization running from US$3.8 to US$16.8 million in the smallest and largest quartiles respectively (Accenture 2017). The picture did not improve in 2018. Data show that over the year 2.6 million people encountered newly discovered malware on a daily basis.1 Attacks ranged over 1.7 million different forms of malware, and 60% of the attacks lasted less than 1 h. Cyber attacks are escalating in frequency, impact, and sophistication. The escalation is due to several factors, for example, attacking in cyberspace is easier than defending; most attacks remain unattributed and, therefore, unpunished. Moreover, as defences are porous, cyber attacks are more likely to succeed than not (Taddeo 2017b). Artificial intelligence (AI)2 could help to improve defences and reduce the impact of cyber attacks. This is why initiatives to develop applications of AI for cybers security applications are attracting increasing attention both within the private and public sector (The 2019 Official Annual Cybercrime Report 2019).