AI agents should not send email with AI-controlled inputs
Why This Matters
AI agents should not have unrestricted ability to send emails with content generated by AI or influenced by user input. Without proper controls, an agent could be manipulated to send unauthorized emails, potentially exfiltrating sensitive data or launching phishing campaigns. IT administrators must ensure that email-sending capabilities are tightly governed to prevent abuse.
What Aether365 Checks
This check scans all Copilot Studio agents configured in your Microsoft 365 tenant for email-sending tools where the recipient, subject, or body field is dynamically controlled by AI-generated content. You will see this finding in the Aether365 dashboard under the microsoft-365 checks section.