Google AI email summaries can be hacked to hide phishing attacks

A security vulnerability in Google Workspace's Gemini AI has been discovered, which could allow cybercriminals to manipulate email summaries and bypass current security measures. The flaw enables attackers to insert invisible commands into email summaries, effectively hiding phishing attempts from users. Researchers have demonstrated that the Gemini AI system, used to generate email summaries, can be exploited to insert malicious code that is not visible in the final summary presented to the user. This could allow attackers to conceal phishing links or other malicious content, potentially tricking users into interacting with these emails. Google has acknowledged the issue and is working on a fix. In the meantime, users are advised to be cautious when relying solely on email summaries and to carefully examine the full email content before taking any action. The discovery highlights the importance of continuous security assessments and the need for robust safeguards in AI-powered systems, as vulnerabilities can be exploited to bypass existing security measures and put users at risk.