Threat
Invitation Is All You Need! Promptware Attacks Against LLM-Powered Assistants in Production Are Practical and Dangerous
First reported · Discovered arxiv.org
Page published
Earliest dated coverage: 3 Jun 2026 · First observed: 25 Oct 2026 · Latest dated coverage: 3 Jun 2026
Coverage timeline
Single-source research — one report is available.
Why it matters
The Gemini demonstrations show how indirect prompt injection can cross from untrusted content into tool execution, putting private data, device applications, and connected home devices at risk.
Google Gemini-powered assistants were demonstrated to follow malicious instructions embedded in calendar invitations, emails, and shared documents, enabling memory poisoning, tool misuse, data exfiltration, and unauthorized device actions. The underlying research demonstrated 14 attack scenarios and reports that Google deployed dedicated mitigations; the supplied evidence does not establish exploitation in real-world attacks.