Phishing attacks target humans and AI assistants via prompt injection

Cybersecurity researchers have identified a new ‘dual-target’ phishing tactic where single emails are designed to deceive both human recipients and their AI-powered email assistants. According to Barracuda Research, these attacks utilize prompt injection to embed hidden instructions within emails—concealed via HTML comments, invisible text, or encoded content—to manipulate how AI summarizes or prioritizes messages.

These malicious instructions can trick AI assistants into flagging fraudulent emails as high priority, instructing users to update vendor payment details, or even manipulating automated CV screening processes. As AI becomes more integrated into business workflows, experts warn that organizations must secure the AI systems that consume email data alongside traditional user inboxes.

Broadly, phishing remains a massive threat. The Anti-Phishing Working Group reported over one million unique phishing sites in the second quarter of 2026, a 10.1% increase from the previous quarter. Attackers continue to use sophisticated methods such as typosquatting and adversary-in-the-middle (AiTM) kits to bypass security measures like multi-factor authentication.