When an Email Talks to Your AI: The Rise of Indirect Prompt Injection
Barracuda Networks, Friday, October 23rd, 2026
Barracuda explains how malicious emails use indirect prompt injection to target AI assistants reading inboxes.
Barracuda says malicious emails are no longer written only for human readers but also for the AI assistants that process inboxes.
Indirect prompt injection hides malicious instructions in content AI is expected to read, including email, documents, images and websites.
Email is a major exposure because it combines untrusted external content with AI access to sensitive data and business tools, and traditional phishing indicators may be absent.
Barracuda recommends layered defenses including least privilege, separating content from commands, human approval, output validation, monitoring and adversarial testing.