Security commitments

Tokens and keys#

• Agent tokens and organization API keys are shown only once, when issued. Atmark can't show them to you again either.
• Lists show only the last four characters.
• If a token leaks, you can rotate or revoke just that one.
• Tokens are only accepted in a header, never in a URL or request body.

The owner's policy#

• Only owners and admins change the policy, in the console. An agent token can't touch the policy, billing, or agent creation.
• New agents start with an empty outbound allowlist. They email no one until you add someone.

Sending#

• Every send, over MCP or REST, passes one gate. The gate checks the suppression list, sending limits, and the owner's policy.
• One emergency stop halts sending for a single agent or the whole organization, right away.
• Decisions and console actions are recorded, and you can review them in the console logs.

Received mail#

• Mail blocked by the inbound policy isn't bounced; it's quarantined, and the agent can't see it.
• Received mail is handed to the agent marked as outside data. See Treat received mail as data.
• Webhooks never carry the sender, subject, or body. Every webhook request is signed.

Identity#

• Public documents never include your organization's name, list contents, numbers such as limits, or mail.
• The agent's private key never reaches the console either.

If you find a security problem, let us know.