Abstract

The evolution of the internet posed several challenges, and cybersecurity became a mandate to ensure the continuity of services and the protection of digital assets. These challenges involved malware and social engineering tactics that made traditional security inefficient. The adoption of AI initially worsened the security posture because AI was considered a double-edged sword (having benefits and drawbacks). However, the adoption of AI and its subdomains (ML, deep learning) has transitioned security from conventional to advanced detection methods. In the past, signature-based detectors were used to detect malware; now ML-based detection systems help to detect malware types and even zero-day vulnerabilities that were too complex for traditional detectors to identify. Moreover, ML-based detection systems are highly capable of identifying threats and mitigating their effects effectively. With time, models also became vulnerable to adversarial attacks, which degrade the performance and reliability of the model. This paper illustrates related papers, highlights current and future challenges, and defense recommendations to build a robust security landscape for the future.