Enforced structurally, not just documented.
Every guardrail on this page lives in the one shared send/ingest path — not duplicated (and possibly forgotten) per API entry point, and not something a caller can route around.
WarmHawk enforces CAN-SPAM (a mailing address per sending domain + unsubscribe), RFC 8058 one-click unsubscribe, an EU AI Act disclosure marker, GDPR erasure, CSV-injection defense, a bounce/complaint circuit breaker (a campaign pauses past 5% on 20 sends, a mailbox past 3% of its last 7 days on 50), login brute-force throttling, continuous blocklist monitoring, and AES-256-GCM credential encryption — all structurally, in the API itself, not as optional dashboard settings.
A send is refused before it reaches the pipeline if the sending domain has no physical mailing address, or there is no unsubscribe link to put in the footer — both are unconditional, structural checks in the one shared send path, not per-entry-point duplicated logic a caller could route around.
The footer address belongs to the domain a mailbox sends from, so an agency’s client brands each print their own address and never share one. There is no install-wide fallback. Adding a domain never needs an address; launching a campaign that sends from it does, and the launch check names every domain still missing one.
A campaign with no unsubscribeUrlTemplate of its own links every email to a page WarmHawk serves on your install’s domain, at /unsubscribe/ plus a signed token that names the lead without putting the address in the URL. Opening the page changes nothing, because mail scanners follow links; pressing the button suppresses the address in every campaign. A suppressed address is also refused at send time, so a lead who unsubscribes with an email already queued does not get it.
List-Unsubscribe and List-Unsubscribe-Post: List-Unsubscribe=One-Click headers are generated server-side on every send, letting Gmail and Yahoo unsubscribe a recipient with a single POST, no login required — never left to template configuration.
A disclosure marker is auto-appended whenever a send is genuinely AI-generated AND the recipient resolves to an EU-region signal. A fallback send (AI unavailable, using the plain template) has nothing to disclose, so the marker only appears on real AI-written content.
DELETE /v1/leads/erase anonymizes a data subject’s PII across every campaign by email, preserving aggregate counts; DELETE /v1/leads/:id is a real hard delete of one row. Both are structural rights, not admin-only tooling.
A lead with a value starting with =, +, -, or @ is rejected on ingest (CSV import lists it under rejected with reason csv_injection_risk; the webhook refuses it), so it is never stored and a malicious lead field can’t turn into a spreadsheet formula when you later export and open the data in Excel or Sheets. A plain @handle or scoped package name like @acme/tools is allowed.
A campaign auto-pauses (pausedForBounceRate: true) once its bounce rate exceeds 5%, after at least 20 sends. A mailbox pauses once more than 3% of its last 7 days of sends bounce, after at least 50 sends in that window — enough to be a real signal, not one unlucky bounce. A mailbox that gets a temporary refusal runs at half pace for a day. Catches a bad list before it damages a domain’s reputation, not after.
POST /v1/auth/login is both rate-limited (10 requests/minute) and per-account locked out after repeated failures — two independent layers, since rate limiting alone doesn’t stop a slow, patient attacker targeting one account.
Every domain check runs against Spamhaus ZEN/DBL, Barracuda, and SORBS in addition to SPF/DKIM/DMARC — a one-time check at setup doesn’t catch a domain landing on a blocklist days later.
Mailbox SMTP/IMAP passwords, OAuth refresh tokens, and BYOK AI provider keys are all AES-256-GCM encrypted server-side before persisting, and never echoed back in any API response — not even to the authenticated caller who just set them.
Rate limits
| Route | Limit |
|---|---|
| POST /v1/auth/login | 10 requests / minute |
| POST /v1/leads/import | 10 requests / minute |
| POST /v1/leads/webhook | 30 requests / minute |
| Public domain-check tool | 20 requests / minute |
| Every other route | 100 requests / minute (global default) |
The send-pipeline guardrails on this page are all inside warmhawk-core-engine. For the licensing/billing security model — RSA-signed licenses, Stripe webhook verification — see Stripe checkout & webhooks. For the per-guardrail request/response detail, see Leads & enrichment and Sending safely & domain health.
Questions
Guardrails & compliance: questions worth answering up front
Are these guardrails enforced by the dashboard, or by the API itself?+
The API itself, in the one shared send path — the dashboard (Tier 1/2) is a UI on top of the same engine, not a separate enforcement layer. A direct API call gets exactly the same guardrails as a dashboard-triggered send.
Can I raise the bounce circuit breaker’s 5% threshold?+
The campaign’s, yes — it’s configurable per campaign via bounceRateThreshold on PATCH /v1/campaigns/:id. The 20-send minimum sample size is not currently configurable, and neither is the mailbox breaker (3% of the last 7 days, 50-send minimum), since one mailbox sends for every campaign it’s on.
What happens to a lead’s data if I never call the erasure endpoint?+
It stays as-is indefinitely — WarmHawk doesn’t auto-expire lead data on any schedule. Erasure (DELETE /v1/leads/erase) is something you call when a data subject makes a request, not a background job.
Where does license/billing security fit into this list?+
Separately — RSA-signed license issuance and Stripe webhook verification are covered in the checkout/billing docs, not here. This page is specifically about sending-pipeline and data-handling guardrails inside warmhawk-core-engine.