Artificial Intelligence in Cybersecurity

Digital transformation is increasing the complexity of IT environments, distributing sensitive data across multiple clouds, and worsening the skills shortage—leaving even well-staffed security teams struggling to keep up with evolving cyber risk. This pressure, combined with highly visible AI successes in other domains, fuels the belief that AI will soon replace humans in “less creative” jobs, including cybersecurity. That expectation requires a reality check.

A key distinction is drawn between Strong AI (“thinking like a human,” largely philosophical) and Artificial General Intelligence (AGI), which aims to perform intellectual tasks broadly like a human and remains an active research field. Most real-world deployments, however, are Narrow AI, trained for specific tasks such as language processing or image recognition. While machine learning’s foundations date to the 1940s, recent growth in cloud computing and specialized hardware has accelerated adoption and enabled thousands of startups to build AI products, ranging from everyday assistants to emerging technologies like driverless cars.

In cybersecurity, AI is positioned primarily as augmentation rather than replacement: automating repetitive work, enhancing visibility, supporting analysts, and accelerating response. Major blockers include limited access to high-quality training data (often too sensitive to share), the difficulty of verifying ML behavior under real-world and adversarial conditions, and the need for continuous retraining as threats and vulnerabilities evolve. The market is expanding with AI-driven analytics (next-gen SIEM/security intelligence) that reduce noise and prioritize context-rich alerts. More “cognitive” approaches (semantic reasoning, threat intelligence extraction, phishing defense) remain immature. “Autonomous” security claims should be treated skeptically: even advanced systems still require human experts and are often delivered as managed services.

• Digital transformation is increasing IT infrastructure complexity and distributing sensitive data across multiple clouds.
• There is an increasing shortage of skilled people to manage modern cybersecurity challenges.
• Strong AI is described as “thinking like a human” and remains largely a philosophical concept.
• Artificial General Intelligence is presented as an active research field with dozens of companies and academic institutions involved.
• Narrow AI is characterized as solving a single, narrow task such as language processing or image recognition.
• The theoretical foundations of machine learning date back to the 1940s.
• Cloud services and specialized hardware recently enabled a surge in accessible computing power for machine learning.
• Cybersecurity training data commonly includes intellectual property, personally identifiable information, and other regulated data.
• Next-generation SIEM/security intelligence platforms reduce false positives and produce context-enriched alerts ranked by risk scores.
• Some “autonomous” AI security products are offered as part of managed service packages because human intervention is still needed.

• “Having AI as potential replacement for overworked security experts… would be awesome, wouldn’t it? Alas, people waiting for solutions like that need a reality check.”
• “First, artificial intelligence, at least in its practical definition, was never intended to replace humans, but rather to augment their powers…”
• “Upgrading to AI-powered tools… may feel like switching from pen and paper to a computer, but both just provide humans with better, more convenient tools…”
• “Formal verification and testing of machine learning models is a massive challenge of its own.”
• “The field of cybersecurity is constantly evolving… cyberthreat models must be continuously updated, expanded and retrained…”
• “Perhaps the most popular approach is applying Big Data Analytics methods to raw security data for detecting patterns or anomalies…”
• “Alas, we are yet to see sufficiently mature products of this kind on the market.”
• “Customers should take such claims with a pinch of salt.”
• “Letting a machine learning model make autonomous security decisions may cause numerous additional legal or ethical implications…”
• “Ignoring their growing influence… and hoping to ‘wait out the hype’ cannot be a sensible long-term strategy.”