SafeBreach found a now-fixed Gemini Android flaw that let malicious WhatsApp and Slack alerts manipulate AI responses and tools.