How an Automated AI Check Reads Every Email: Safety Without Latency

Most email platforms wait until after severe abuse reports or spam blacklistings before looking at account activity. Others try to run heavy artificial intelligence models before every send, making API calls take seconds instead of milliseconds. email59 is one of the first email services to use a two-tier compliance system: instant preflight filtering plus an automated background AI audit.

Every email sent by a customer and every email delivered into a mailbox is reviewed by an automated AI check. Here is how it keeps the platform kid-safe and non-marketing without slowing down your application.

Tier 1: Instant preflight checks

When you call POST /v1/send or compose a message in your mailbox, the preflight send policy checks your text in microseconds:

• Zero API latency. The send check uses fast pattern matching rather than waiting on an external language model. Your API calls return in standard network time.
• Blatant violations are blocked up front. Clear marketing phrases and kid-safe filter triggers are stopped immediately with HTTP 422 before the email leaves your app.
• Clean transactional mail goes straight to delivery. Verified password resets, receipts, and alerts are dispatched instantly.

Tier 2: Background AI compliance reading

After an email is processed, an automated background worker reads the message content to ensure it complies with email59's kid-safe, non-marketing policy. The AI reader evaluates the subject and body and returns one of four structured verdicts:

• ok: Ordinary personal correspondence, school messages, order receipts, sign-in codes, and support updates.
• not_kid_safe: Profanity, sexual content, graphic violence, hate speech, illegal substances, self-harm, gambling, or phishing scams.
• marketing: Unsolicited promotions, sales pitches, discount codes, newsletter blasts, or commercial solicitation.
• other: Policy issues that fall outside the categories above.

What happens when a violation is found?

When the background AI reviewer detects that a customer sent mail violating our policies, the platform acts automatically to protect recipients:

• Sending is immediately held. The sender's account flag send_hold is set.
• All sending routes are locked. Calls to POST /v1/send, the MCP send_email tool, the Supabase hook, and web mailbox compose immediately return HTTP 403.
• No further messages can be sent. The account cannot send any additional emails until reviewed by human support.

Here is the exact response an account on hold receives:

HTTP/1.1 403 Forbidden
Content-Type: application/json

{
  "detail": {
    "error": "sending_on_hold",
    "message": "Sending is on hold. Contact support@email59.com."
  }
}

Human review and resolution

If your account sending is placed on hold, reaching out to support@email59.com initiates a manual review:

• False positives can be cleared. If a benign transactional message was flagged incorrectly, support can clear the hold and tune the rules.
• Repeated or deliberate abuse is blocked permanently. Accounts created to blast spam or bypass child safety protections are terminated to preserve the integrity of our sending network.

A cleaner, safer email network

Automated AI reading ensures that email59 remains a safe haven for real human conversation and essential application events.

Read the docs Read our safety terms