A critical security flaw in Google Gemini, the AI tool embedded in Google Workspace apps like Gmail and Docs, has been revealed. This vulnerability permits attackers to embed invisible commands in emails, transforming the AI into an unwitting participant in sophisticated phishing schemes. The issue arises when Gemini’s “Summarize this email” feature processes hidden code as legitimate instructions, leading to convincing, fabricated security alerts that can deceive users into disclosing sensitive information.
Researchers warn that this threat extends beyond email, potentially affecting all Google Workspace applications, increasing the possible attack surfaces. Users are at heightened risk for credential theft due to these AI-driven deceptions, which can occur without any user error, making them difficult to detect.
To combat these threats, it’s recommended to implement inbound HTML linting, configure LLM firewall settings, and enhance user training. As AI becomes integral to digital life, securing these technologies is essential to thwart sophisticated manipulations.
Source link