Malicious Notifications Could Trick Google Gemini Users

SafeBreach researchers disclosed a novel prompt-injection technique called Fake Context Alignment that abused Google Gemini's notification summarization to hide malicious instructions (using muted hyperlinks and foreign/invisible text) and trigger unauthorized actions via delayed tool invocation; the technique could enable device control, unauthorized streaming, social-engineering impersonation, and poisoning of LLM memory. The issue was responsibly reported and Google rolled out classifier updates to mitigate it, and there is no evidence of in-the-wild exploitation at the time of the report.