The attack relies on hidden prompts in a foreign language.
SafeBreach researchers showed how hidden commands in Android notifications could trick Google Gemini through indirect prompt ...