A flaw in Google Gemini has been exploited by criminals for phishing attacks, utilizing summarized emails to launch deceptive campaigns despite Google being aware of the issue since last year. Although it hasn’t been addressed yet, cybersecurity experts highlight this ongoing vulnerability.

Embedding Invisible Text

Criminals can implant hidden messages within emails through HTML tricks like white text or concealed formatting. When an email appears harmless to the recipient, Gemini’s AI reads and processes everything, including invisible content. If requested for a summary by the user, Gemini may inadvertently include this hidden text in its output.

Phishing Traps

The summary generated by Gemini might contain warnings about compromised passwords or other urgent alerts that look legitimate since they originate from the AI itself. Recipients are thus more likely to trust and act on these messages, such as changing a password or contacting a support number.

Evading Filters

Apart from hidden text, criminals omit suspicious links or attachments to bypass Google’s spam filters, allowing the malicious emails to reach inboxes undetected. This method leverages AI for polymorphic phishing techniques that randomize components of fraudulent emails, thereby evading detection systems.

Technical Challenges

Detecting these hidden threats is technically complex. Filters might monitor Gemini’s output for urgent messages or suspicious links but face challenges in identifying concealed content. Education remains a crucial defense against such sophisticated phishing attempts.

AI’s Double-Edged Sword

This isn’t the first instance of AI being co-opted for phishing. Polymorphic phishing uses AI to randomize aspects of emails, making them harder to detect. Interestingly, Google has highlighted Gemini’s role in cybersecurity efforts and its integration into platforms like Google Threat Intelligence.