Mailbuttons The owner has verified their profile information displayed on this page.

Questions and answers

• How do your server-side policies handle forwarded emails, spoofed sender addresses, and messages with malicious content hidden in attachments or quoted text?

  Asked by SaaS Browser · November 2, 2026

  Answered by Richard Halldearn from Mailbuttons · November 3, 2026

  Spoofed sender addresses. Every inbound message is authenticated at the SMTP edge by our Stalwart mail cluster, which runs SPF, DKIM and DMARC checks and stamps the result into an Authentication-Results header. Our policy engine reads only the stamp written by our own server, matched by its authserv-id, so a sender cannot forge a "pass" by including their own Authentication-Results header. Each mailbox carries a server-side MailPolicy with an ordered sender allowlist by address or domain, and any rule can set requireDkim or requireSpf. A sender that matches no rule is bounced or dropped by default, and a matched sender whose signature fails is rejected before any agent code or model sees the message. New mailboxes are seeded fail-closed: the owner's address is allowed, and everyone else lands under a passive, DKIM-required catch-all that files the mail and never invokes the model. Rejection notices go only to senders with a DMARC pass and an aligned From header, are never sent to automated senders, and are rate-limited per sender, so we do not generate backscatter to forged addresses.
  Forwarded emails. In Live and Supervised modes only our own edge verdict counts. Because DKIM normally survives forwarding and SPF does not, the default catch-all requires DKIM alone, so legitimately relayed mail is not bounced while spoofed mail still fails. The one relaxation is Shadow mode, used for inbox assessments where a customer forwards their real traffic to us. There, if our edge verdict fails because a forwarder rewrote the body, we may accept the upstream tenant's own DKIM or SPF stamp as evidence. That path is structurally unreachable outside Shadow mode, and a Shadow mailbox can never send, act, or bounce, so weaker evidence can never trigger a real action. The audit row records which evidence source was used.
  Malicious content in attachments or quoted text. Policy evaluation runs server-side on the plain-text body before the model is invoked. Content guards are regex rules that reject a message outright, with the reason recorded in the audit log. They run over the whole text part, including quoted reply trails, since quoted text is part of the body. HTML parts are not rendered to the model. Attachments are never fed to an agent automatically. An agent must request them explicitly, and the request requires an attachments capability on its token, is capped at 25 MB, only renders text, CSV and PDF as text, and runs PDF extraction in a panic-isolated path because attachments are treated as untrusted input.

• Audit trail
  Keeps a clear record of all email activity for security and compliance.
• High availability
  Guarantees 99.9% uptime for reliability.
• Sender allowlists
  Allows only approved senders to send emails, ensuring safety.
• Developer-friendly
  Offers easy options for evaluation and pricing for developers and teams.
• Enterprise options
  Provides custom solutions for larger, regulated companies.
• Policy enforcement
  Runs email rules like DKIM, SPF, and allowlists on the server before AI sees the message.
• Auditable by default
  Automatically records details of each email for compliance and analysis.
• EU/UK data residency
  Stores data in EU and UK regions to follow local laws.
• Exportable audit logs
  Allows exporting logs to other security tools like SIEM.
• Standards-compliant JMAP
  Uses open standards for email, avoiding special lock-in.
• Full-duplex JMAP protocol
  Supports sending and receiving emails efficiently through a standard protocol.
• Security review readiness
  Designed for environments that need security checks before use.
• Tamper-evident audit logs
  Logs cannot be changed, showing any tampering or illegal changes.
• Standards and certifications
  Uses recognized standards like ISO 27001 for security.
• UK-incorporated and EU/UK-hosted
  Based in the UK and EU to meet local business needs.
• Email infrastructure for AI agents
  Provides email addresses and tracking for AI agents to send and receive emails securely.
• Protection against prompt injection
  Prevents unwanted commands from being sent to AI via email.