Attackers Hide AI Prompt Injections Inside Phishing Emails
Barracuda found phishing emails hiding prompt injections aimed at AI inbox assistants and human recipients.
Barracuda said on October 7 that it analyzed phishing emails carrying hidden prompt injections for AI assistants alongside conventional lures for recipients. Samples used matching sender and recipient addresses, a trusted spam score, and a public-sector domain, plus password-protected attachments with the password written in the body. Instructions were concealed with HTML comments, CSS-invisible text, Base64, and zero-width characters, including an invoice prompt that tried to add a fake payment-change action. Barracuda did not report the campaign's scale and recommended stripping hidden content, sandboxing AI, validating outputs, and requiring human approval for payments.
- Single emails target people and AI inbox summarizers together.
- Password-protected attachments included passwords in the message body.
- Prompts hid in HTML comments, CSS, Base64, and zero-width text.
- An invoice sample nudged an AI summary to redirect payments.
- Barracuda did not say how widespread the campaign was.
From the original article
2 min read at infosecurity-magazine.com
Barracuda finds phishing emails designed to manipulate both human users and AI assistants
By Alessandro Mascellino · infosecurity-magazine.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.