How Attackers Are Using AI, and How Defenders Are Fighting Back
Every significant technology eventually finds its way into the hands of attackers, and artificial intelligence is no exception. Over the past several years I have watched the discussion evolve from speculation about what AI-enabled attacks might look like to documented evidence of how adversaries are actually using these tools. The picture that emerges is less cinematic than some predicted, but it is consequential nonetheless. Fortunately, defenders are adopting AI at least as rapidly.
How Attackers Are Using AI Today
The most widespread use is in social engineering. Generative AI produces fluent, convincing phishing messages in any language, free of the grammatical errors that once served as warning signs. It allows attackers to personalize messages at scale, drawing on publicly available information about targets. Voice cloning and synthetic video have enabled impersonation of executives in fraud schemes, including several well-publicized cases involving fraudulent payment authorizations.
AI also accelerates attacker productivity more broadly. It assists in writing and modifying malicious code, researching targets, analyzing stolen data, and automating reconnaissance. In my assessment, the principal effect so far is not the creation of entirely new attack categories but the lowering of skill barriers and the increase of speed and volume.
A newer concern involves attacks on AI systems themselves: prompt injection against enterprise AI applications, manipulation of AI agents, and poisoning of training data.
How Defenders Are Responding
Defenders hold an important advantage: access to vast quantities of security data and the ability to train models on it. AI is now central to several defensive capabilities. Behavioral detection identifies anomalies in user, device, and network activity that suggest compromise, regardless of how convincing the initial phishing message was. Inline analysis classifies previously unseen malware, phishing pages, and malicious domains in real time. Correlation and triage reduce alert volumes and surface genuine incidents for analysts. Automated response contains threats at machine speed. And AI copilots help analysts investigate faster and with greater confidence.
Human Defenses Still Matter
Technology alone will not defeat AI-enhanced social engineering. Organizations should update training to reflect convincing AI-generated messages and deepfakes, and establish verification procedures, such as call-back protocols using known numbers, for payment requests and sensitive changes.
The Bottom Line
Attackers are using AI to craft more convincing lures, impersonate trusted individuals, and operate with greater speed and scale. Defenders are responding with AI-driven detection, real-time prevention, automated response, and analyst assistance, supported by updated training and verification procedures. In my view, the contest will increasingly be one of AI against AI, decided by the quality of data and the speed of action. Organizations that invest in AI-powered defenses now will be best positioned as adversaries continue to adopt these tools.