An email can contain instructions aimed at an AI assistant rather than at you. For example, a message may tell the assistant to ignore the user’s request, reveal unrelated messages, or recommend an attacker-controlled destination. That content should be treated as material to inspect, not as authority over your own instructions.
Recognize the suspicious change of role
Look for text that addresses an assistant, system, or tool and tries to redirect its behavior. The message may mix ordinary business information with those directions. Familiar names or accurate project details do not make the embedded instruction trustworthy.
Do not test the message by asking the assistant to obey it. If a response unexpectedly asks you to disclose information or perform an unrelated action, return to your original task and inspect its sources.
Understand what Google’s protections can and cannot establish
Google says Gemini in Workspace can warn about risky content, exclude some material, or refuse a response when malicious activity is detected. A missing summary may therefore reflect a protective action rather than a broken mailbox.
These controls do not justify assuming every suspicious message is harmless. Keep normal phishing checks in place and verify any request involving credentials, payments, or confidential information through an established channel.
Respond without spreading the content
Use Gmail’s phishing-report workflow for a deceptive message and the available Gemini feedback controls for unsafe assistant behavior. If this concerns a work account, follow the organization’s incident process and preserve the minimum evidence it requires.
Do not forward the suspicious message broadly, copy its instructions into an unrelated AI service, or grant additional file access to reproduce the effect. When describing the problem, record the original task, suspicious instruction, and unexpected response while concealing sensitive content. A source citation can help explain where the instruction came from, but it is not approval to follow it.
Browse Gmail & Google Account help
References
- How Google Workspace with Gemini helps protect users from malicious content and prompt injection
- Learn how to use sources with Google Workspace with Gemini
Reviewed against the linked official product guidance on November 3, 2026. Features, limits, screen labels, and access can vary by account, platform, region, and organization policy.