Cybersecurity and generative AI: new threats and how to defend against them

Generative AI amplifies cyber threats: hyper-personalised phishing, deepfakes, polymorphic malware. Here's how to adapt your defence. Generative AI as an offensive weapon: Generative AI has democratised sophisticated cyberattacks. An attacker can now generate perfectly written, personalised phishing emails in seconds, create audio deepfakes to impersonate executives (CEO fraud), and produce polymorphic malware that evades traditional antivirus. Attack cost has plummeted while quality has soared. AI-augmented phishing: AI-generated phishing campaigns are 60% more effective than traditional ones (IBM 2025 study). AI analyses LinkedIn profiles, publications and previous emails to create hyper-contextualised messages. Classic spam filters are outmatched. Deepfakes and CEO fraud: In 2024, a Hong Kong company lost $25 million after an employee participated in a video call where all participants were deepfakes of colleagues. Solution: implement multi-channel validation procedures for sensitive operations. AI as a defensive shield: Defensive AI analyses network behaviours in real-time (UEBA), detects anomalies invisible to humans, and automates incident response. Tools like CrowdStrike, SentinelOne or open source solutions (Wazuh) now integrate ML models for proactive detection. 5 priority measures: 1) Train teams specifically on AI attacks. 2) Deploy phishing-resistant MFA (FIDO2/WebAuthn). 3) Deploy EDR/XDR with AI behavioural detection. 4) Establish multi-channel validation for sensitive operations. 5) Regularly audit your attack surface with offensive AI tools (red teaming).