Phishing emails are baiting your AI assistant

Phishing emails hide prompt injections that try to make AI inbox assistants endorse fraudulent payment changes.

Malwarebytes reports Barracuda research on phishing emails that target both recipients and AI assistants that summarize inboxes. Messages pair a password-protected attachment and password with hidden instructions meant to make a summarizer treat the email as legitimate or add a fake priority action, such as changing a vendor’s payment details. Concealment includes HTML comments, invisible styling, encoding, and zero-width characters. Defenders are advised to verify money and data requests in the original message, limit assistant permissions, and require human approval for payments.