Limitations of AI in cybersecurity
AI algorithms namely machine learning and deep learning algorithms are powerful tools. However, they suffer from some limitations which require that human analytics should work with AI tools collaboratively. In this post, we will look at the most important shortcoming of Artificial Intelligence in the cybersecurity domain. Though Benefits are more, AI also comprises limits [4].
Here are some of the potential limitations of AI:
Cyber threats are constantly evolving:
Cybercriminals are creative and come up with new ways to conduct cyberattacks. For instance, Phishing emails are evolving rapidly which requires machine learning algorithms to be retrained and updated.
Cybercriminals use AI, too:
They take advantage of machine learning and deep learning algorithms to understand the behavior of AI tools and try to design their attacks based on the vulnerabilities of the model [1].
Vulnerable to Adversarial attacks:
Adversarial examples are inputs to machine learning models that an attacker has intentionally designed to cause the model to make a mistake [2]. Cybercriminals conduct different Adversarial attacks by crafting Adversarial examples in innovative ways. For example, designing samples to fool Phishing AI detection tools to classify phishing examples as normal.
AI FALSE POSITIVES
AI and machine learning are not deterministic. This naturally includes incidents of 'false alarms' and 'false-positives'. In Cybersecurity, a high rate of false alarms makes the product unusable. Therefore, many AI solutions require a process of fine-tuning, specialized configuration, and adjustment for each organization, sometimes at different networks within an organization [3].
Conclusion and Recommendations:
AI is a powerful tool and has acquired significant achievements but, we shouldn't expect it to do everything and solve all problems especially in the cybersecurity domain. If we are planning to consider applying AI techniques in our cybersecurity projects, it is recommended to become familiar with AI limitations.
