Malicious WhatsApp, Slack Alerts Could Have Exposed Millions of Android Users

SafeBreach researchers discovered a notification-based prompt-injection vulnerability in Google Gemini’s Android Utilities that could allow crafted notifications from messaging apps to manipulate how Gemini processes and speaks notification content (a technique called Fake Context Alignment). Google applied server-side content-classifier mitigations, there is no evidence of exploitation in the wild, and users can limit exposure by disabling Gemini Utilities or notification read/reply permissions.