Cyber & IT Supervisory Forum - Additional Resources

ARTIFICIAL INTELLIGENCE AND CYBERSECURITY RESEARCH

AI IN CYBERSECURITY

This section summarises the current state-of-the-art in the main uses of ‘traditional’, long-standing and newer AI applications (deep learning systems), tools and methods in cybersecurity, looking at both sides of the use of AI in the context of cybersecurity requirements, i.e. malicious and virtuous. Below, is a non-exhaustive list of ways AI can be used in cybersecurity: a. Cyber-criminals exploiting AI to boost their efficacy; b. Security mechanisms encompassing AI to detect, identify and mitigate the consequences of compromises; c. Using AI to exploit vulnerabilities in existing AI and non-AI tools and methodologies, e.g. adversarial attacks 61 ; d. Using AI during the design of a system to protect existing AI and non-AI tools and methodologies (protection created during system design). In the first two cases, AI is used as a tool (an attacker can use AI to design the attack), while in the last two cases AI is the actual target (the attack may target an AI-based system). Even though AI-based defence mechanisms address a wide variety of vulnerabilities, they can themselves be points of attack. Attackers use AI not only to orchestrate various cyber threats, but to attack AI-based defence mechanisms by exploiting existing vulnerabilities. The table below identifies the use of AI methods in cybersecurity functions.

Table 3: AI methods in cybersecurity functions (source: authors)

Security function\AI DT SVM NB

K means

HMM GAS ANN CNN RNN Encoders SNN

X X X

X X

X X

X X

X

X

X

X X

X X

X

Intrusion detection Malware detection

Vulnerability assessment Spam filtering

X

X

X

Anomaly detection Malware classification Phishing detection

X

X X

X

X

X

Traffic analysis

X X

Data compression Feature extraction

61 For example, in GANs given a training set, the technique learns to generate new data with the same statistics as the training set. A GAN trained on photographs can generate new photographs that look at least superficially authentic to human observers, having many realistic characteristics.

19