Is Google’s Gemini AI Model for Workspace Susceptible to Phishing Attacks?

A vulnerability in Google’s Gemini AI model for Workspace has been identified, allowing attackers to exploit the “Summarize this email” function to execute hidden malicious prompts. The technique involves embedding commands in email content using HTML and CSS with zero font size and white text, making them invisible to Gmail users. When Gemini summarizes an email, it unknowingly executes these hidden instructions, potentially generating fake warnings about account breaches. This method bypasses traditional phishing detection, increasing the likelihood of users trusting and acting on fraudulent notifications. This innovative attack exploits users’ trust in AI, making them more susceptible to phishing attempts.