Lock quarantined mail away from your agent
Hi everyone 👋
When we launched, we gave agents a lot of credit. Quarantined mail was withheld by default, but an agent could still ask for the raw text if it decided it needed it. Our thinking was that a well-behaved agent would leave it alone unless a person asked.
After a few rounds of feedback with users, we learned that this is not enough for everyone. Some of you told us you need quarantined text locked down completely: no flag, no fallback, no way for the agent to retrieve it.
So we shipped the agent-safe view. Create a key with the new "Agent (safe view)" preset in the dashboard, or grant the `message_safe_read` permission through the API, and that key:
reads the cleaned text of ordinary mail, without raw HTML, headers or text hidden from human readers
sees only the sender's address of a quarantined message, with no subject, body, attachment names or links
cannot fetch the raw message or the inbox export, forward quarantined mail, or remove the quarantine label
gets the same view in events, webhooks and the MCP tools, where include_raw is refused
Existing keys are unchanged. A person can still open quarantined mail in the dashboard. New sandbox agents now start in the agent-safe view too, until an organization claims them.
Changelog: https://machinemail.dev/changelog/agent-safe-view
Docs: https://machinemail.dev/docs/concepts/api-keys#the-agent-safe-view
If you run agents on real inboxes, we would like to hear how you decide what an agent may read. What else should it never be able to reach?


Replies
Be the first to reply
Have a question or a thought to share? Add a comment above to start the conversation.