A plain-language summary for customers, prospects and their security teams. No legal language, no claim we have not implemented.
The one thing that makes DMARC Harbor different§ Link
We do not store your recipients' real email addresses unless you explicitly ask us to.
When a mailbox provider sends a forensic DMARC report, it includes the email address of everyone who received a message that failed authentication. Almost every DMARC tool stores that address in plain text.
We hash it with a one-way HMAC by default. The stored value looks like ps1:7f3a9c2e.... It cannot be reversed into an email address. It is not your recipients' personal data, and it is not ours to hold.
If you need the real address (to chase a specific delivery problem, for example) you switch named mode on for that domain. That switch requires you to confirm a legal basis, encrypts the values with AES-256-GCM, and expires them after 7 days regardless of plan. No tier of our product extends that.
You can switch named mode off at any time, which permanently destroys the encrypted values. We keep the surrounding evidence: which sender failed, how many messages, when. That contains no recipient data.
Our domains§ Link
We use exactly two domains, and their subdomains:
- dmarcharbor.com, for the website, the app, our email and the report mailboxes (
reports.dmarcharbor.com). - dmarcharbour.com (British spelling), for the records we host for you: hosted DMARC, SPF, BIMI and MTA-STS records point at names under it. Opening it in a browser takes you to dmarcharbor.com.
Mail, links or DNS instructions that claim to be from DMARC Harbor on any other domain are not from us. Forward them to security@dmarcharbor.com.
Access control§ Link
| Role | Clients | Domains | Reports | Forensic evidence | Named recipients | People | Billing |
|---|---|---|---|---|---|---|---|
| Owner | Add, change, delete | Add, change, delete | Upload, change, delete | Read, upload, purge | Yes | Invite, change roles, remove, hand on ownership | Read and change plan |
| Admin | Add, change, delete | Add, change, delete | Upload, change, delete | Read, upload, purge | Yes | Invite, change roles, remove | Read only |
| Analyst | Read only | Read and change, not add or delete | Upload and change, not delete | Read and upload, recipients hidden | No | No | Read only |
| Viewer | Read only | Read only | Read only | No | No | No | Read only |
| Portal contact | Their own client only, read only | Read only | Read only | No | No | No | No |
A portal contact is a client's own person, added under Client portal access. They see only the clients they were given, and nothing about the agency's other clients, its team or its billing.
Each workspace has exactly one owner. The plan, the workspace allowance and the right to delete the workspace come from the owner's account, so nobody can be made a second owner. The owner can hand the workspace to another member under Members: they confirm with their password or an authenticator code, the member becomes the owner and the previous owner becomes an admin in one step, both are emailed, and it is recorded in the audit trail. It is refused if the new owner's plan has no room for another workspace, for the workspace's clients and active domains, or for its team (members plus open invitations).
Permissions are individual statements, not a single flag, so "can view reports" and "can see who failed" are genuinely separate decisions. Being able to read reports does not give anyone the ability to see recipient addresses.
Only the owner can change the plan. That is deliberately separated from who can read billing, so a compromised analyst account cannot change what the company is paying for.
Sign-in security§ Link
- Email must be verified before sign-in
- Sessions last 7 days
- You can see every device you're signed in on: device, location, when it started, and end any of them
- You can sign out everything else in one click, or everything including your current device
- Changing your password signs out every other device, unless you untick "Sign out my other devices" on Account security. The box is ticked by default, and the server signs the others out whenever a request does not say otherwise. Resetting a forgotten password by email always signs out every device, including the one you are on
- Setting your new password to your old one is rejected
- Two-step verification with an authenticator app and one-time backup codes, on every plan. A workspace owner can require it for every member, and a session that signed in without it is refused there
- Single sign-on with SAML or OIDC on Admiralty. Just-in-time provisioning admits only the email domains a workspace allows, and never grants the owner role
- Sign-in on an agency's own domain is accepted only from a domain the agency has proved it controls (a verified DNS record) and whose plan includes it. The request must arrive through our web app, which says which host the browser asked for; the same
Originsent straight to the API is refused, as is any other website. Production accepts https only - Password reset and verification links go to an agency's domain only for that agency's client contacts who hold no seat in any workspace. Those contacts sign in on the agency's domain anyway, so the link there exposes nothing the agency's own sign-in page does not. Anyone else who starts a reset there, the agency's own team included, gets a link to our own address, so a token for a staff account is never sent to a host the agency points
Client mail sent in an agency's name§ Link
- Reply-To is only the agency's published support address. An agency can set a support email under its white label settings, and that address, and nothing else, becomes the Reply-To on its client mail. It is never filled in from a staff member's sign-in address. With none set there is no Reply-To. The address is checked as one plain address with no line breaks before it is saved, and again before it is put in a header
- An agency's own sending domain (Admiralty) is added to our email provider's account and later removed from it, so a domain we send from or serve, any subdomain of one, and any parent of one are refused: the domains of our From addresses and report mailboxes, and the hosts of our app, API, web origins and hosted DNS. An agency cannot have us add or remove an entry our own mail depends on. Mail uses the agency's address only after the provider reports its domain verified; if the provider later refuses it, the message is sent once more from the shared address with the agency's name
- The provider key for domains can be separate.
RESEND_DOMAINS_API_KEY, a full-access key, is used only to add, check and remove sending domains, so the key every message is sent with can stay sending-only
Tenant isolation§ Link
Every single query is scoped to a workspace. There is no combined view across workspaces available to any user, including our own staff.
On an agency's own verified domain the client portal shows only that agency's reports. A contact whom two agencies serve is not offered the other agency there, cannot open it by naming its workspace, and a contact the agency does not serve is refused with a message that names no other agency. The host is taken from headers our web app sets; a caller who forges them can only narrow what a contact sees, never widen it, because the contact's own grants still decide what can be shown.
We maintain one authoritative inventory of which data belongs to which workspace, client or domain. The export feature and the erasure feature both read that same inventory, which is what makes it structurally impossible for us to export something we would not erase, or to fail to erase something we told a customer we hold.
Cross-workspace access attempts are covered by automated tests.
Credentials you give us§ Link
Cloudflare tokens, PSA credentials, report mailbox passwords, single sign-on client secrets, webhook signing secrets, and the Slack and Teams addresses alerts post to are encrypted at rest with AES-256-GCM. None is ever shown again in full, and none is included in a data export.
Outbound calls to addresses you supply (webhooks, your PSA, your mailbox host, a logo URL) are refused when they resolve to a private or reserved network address. Webhook, PSA and logo requests then connect only to the address that was checked, so a DNS answer that changes between the check and the call cannot redirect them.
Data intake§ Link
DMARC reports arrive from mailbox providers. Ingestion is authenticated:
- A shared secret signs the report with an HMAC over a timestamp and the raw message
- The signature is compared in constant time, so it cannot be guessed by timing
- Reports older or newer than a 300 second window are rejected
- Raw email bodies are never stored
Report volume is rate limited, and the service rejects oversized payloads.
Audit trail§ Link
Every privileged action is recorded in an append-only log:
- Invitations sent, sent again, withdrawn, accepted and declined; role changes, with the role before and after; people removed or leaving; and ownership transfers
- Clients and domains added, one at a time or in bulk, and each domain's ownership first verified, whether by a person, an API key or the background check
- Plan changes, the billing currency chosen, and entitlement overrides
- Forensic data purges, identity toggles, and forensic collection turned on or off
- Report share link creation and revocation
- Session revocations, two-step verification changes and single sign-on events
- Every DNS change made for you: records published through Cloudflare, and every change to a hosted DMARC, SPF, BIMI or MTA-STS record, each with the value it replaced
- Policy approval requests, and the client's approval or decline under their own account
- API keys, webhooks, PSA, Slack and Teams connections created, changed, paused, resumed or removed
- Data exports requested, downloaded and revoked, and monthly usage exports downloaded
- Data erasures requested, cancelled and completed
Each entry records who, when, what, and a request identifier for correlation with your infrastructure logs.
When you erase data, the audit trail is anonymised rather than deleted. The action, time and target survive so we can prove the change was controlled; the person, IP address and request identifier are cleared. Deleting the trail would leave no evidence the erasure happened.
What we never export§ Link
A data export never includes:
- Passwords
- Session tokens
- OAuth tokens
These are not your data to download: they are *the means by which the download is authorised*. Exporting them would hand over the ability to impersonate the account. You still get the device, address and expiry of each session so you can see where you are signed in.
Where we do withhold something, the export file tells you what it withheld and why, so it can never be mistaken for a complete copy of storage.
Detection§ Link
- New sending source detection. Sources first seen in the last seven days are flagged, and graded by whether they passed SPF, DKIM, both, or neither. A source that passes neither is surfaced as a possible spoofing attempt, because a legitimate service normally passes at least one.
- Anomalous authentication raises an email alert rather than waiting for someone to open a dashboard.
- Alerts support acknowledgement, escalation, owner roll-up, and quiet hours in each recipient's timezone.
Engineering practice§ Link
- Type checking, linting and a build step gate every change
- Over 1,000 automated tests, run on every change, covering authorisation, tenant isolation, crypto, input limits, and every destructive operation's scope
- Dependency audit currently reports zero known vulnerabilities
- The database schema is version controlled with reviewed migrations
- Secrets live in environment configuration, never in source. The service refuses to start in production with development secrets or a development email provider, so a misconfiguration fails closed
What we have not yet done§ Link
Being straight about the gaps, because a security page that only lists strengths is not worth reading:
| Gap | Status |
|---|---|
| SOC 2 Type II | Not started. A report can only be issued by an independent licensed auditor after an observation period |
| Penetration test | Not yet commissioned |
| Hosting platform, backup schedule, recovery objectives | Not finalised |
| ISO 27001 | Not started |
Reporting a vulnerability§ Link
We would like to hear about security problems.
Email security@dmarcharbor.com. Do not include customer data in a report. We will acknowledge reports within 3 business days. The same address is published in /.well-known/security.txt.
Documents§ Link
| Document | Purpose |
|---|---|
docs/PRIVACY-POLICY.md | What we collect, why, and how long we keep it |
docs/DPA.md | Article 28 Data Processing Agreement |
docs/SUB-PROCESSORS.md | Every third party that touches your data |
docs/SOC2-READINESS.md | Compliance track and the controls already implemented |