Jan 21, 2026
Google AI csam gemini
’s for Workspace has been shown vulnerable to “indirect prompt injection” that allows hidden HTML/CSS instructions embedded in email bodies to manipulate Gemini’s summaries and push fraudulent securit...
Your fact-checks will appear here
Other researchers expanded the threat model to calendar exfiltration and misleading event creation, illustrating that natural-language APIs and assistant features can be manipulated to leak or misrepresent private data.