How do I prevent AI agents from sending unauthorized emails?

AI agents with email capabilities can send thousands of messages in seconds — spam campaigns, phishing emails, data exfiltration via email, or legitimate-looking messages that damage your brand — and most agent frameworks provide no controls over outbound communications.

Email-related risks from AI agents:

• Bulk spam: Agent sends promotional emails to thousands of recipients, violating CAN-SPAM and destroying sender reputation
• Data exfiltration: Agent attaches sensitive documents to outbound emails sent to attacker-controlled addresses
• Phishing: Agent generates convincing phishing emails using your company's domain and templates
• Brand damage: Agent sends inappropriate or incorrect responses to customers using your official email

Exogram's Gate 3 (Transmission Guard) enforces a simple rule: agents draft messages — humans send them. All direct email.send, message.send, and sms.send operations are blocked at the execution boundary. The agent can compose, suggest, and queue messages. But the send button requires human approval. This single rule eliminates every email-based risk vector.