Four kinds of email attack
The chapter opens with a message to an accounts-payable clerk that appeared to come from Meridian's CFO and asked for a supplier's bank details to be changed. The clerk reported it instead of acting. The book sorts email threats into four shapes: spoofing of your own domain, phishing and spear-phishing (often from lookalike domains), malware delivered by attachment or link, and business email compromise (BEC), a plausible request for money or data with nothing technically wrong in it.
Each has its own defence. Authentication stops spoofing; filtering, impersonation protection and trained users reduce phishing; sandboxing stops malware; and BEC, which can even come from a genuinely compromised internal mailbox that passes every check, needs process controls such as a second signatory for payment changes. Email security is described as adding evidence to an open channel rather than closing it.
SPF and DKIM
SPF is a TXT record listing the servers allowed to send for your domain. It should end in -all (hard fail) once all legitimate senders are known; ~all (soft fail) is a stage on the way. SPF has limits: it checks the hidden envelope sender rather than the visible From address, it breaks when mail is forwarded, and it allows only ten DNS lookups, so the list of included services should stay lean. Domains and subdomains that never send mail should publish v=spf1 -all and a DMARC record of p=reject so they cannot be borrowed.
DKIM adds a cryptographic signature to each outgoing message, with the public key published in DNS under a selector. For Microsoft 365 this usually means two CNAME records and then enabling signing in the admin centre; two selectors allow key rotation without interruption. DKIM survives forwarding, but on its own it does not tie the signature to the visible From address or tell receivers what to do on failure. A common gap is a transactional sender on a subdomain that was never signed.
DMARC as a staged rollout
DMARC requires that the domain authenticated by SPF or DKIM aligns with the visible From address, publishes a policy for failing mail, and asks receivers to send aggregate reports. The book insists that DMARC is a rollout, not a switch. Start with p=none and an rua address, read the reports through a viewer for several weeks, and fix every legitimate sender. Then move to p=quarantine, ramping with the pct tag, and only after that runs clean set p=reject.
At Meridian, Omar Haddad spent three weeks at p=none, found four sources (the tenant, a marketing platform missing from SPF, a loyalty-app mail service without DKIM, and spoof attempts), fixed the legitimate ones, ramped quarantine over a month and reached reject in about six weeks without incident. Enforcement stops exact-domain spoofing but has no authority over lookalike domains. An expert section covers relaxed versus strict alignment, the sp tag for subdomains, why forwarding shows SPF failures that DKIM rescues, and why forensic (ruf) reports should not be relied on.
Filtering, impersonation protection and malware defence
In Microsoft 365, Exchange Online Protection and Defender for Office 365 cover what authentication cannot. Impersonation protection should list the people most likely to be impersonated and your own domains, enable mailbox intelligence, and quarantine or warn rather than deliver silently; the external-sender tag makes an outside message claiming to be an executive visibly inconsistent. Safe Attachments detonates files in a sandbox, and Safe Links checks URLs at click time, which matters because attackers often weaponise a link after delivery.
Two further decisions: disable automatic forwarding to external addresses by default and alert on new forwarding rules, a common post-compromise trick; and honour other senders' DMARC policies on inbound mail. Quarantine needs an owner, a review routine and a safe way for users to request release, or users lose trust in filtering.
People as sensors
A carefully written spear-phish from a new lookalike domain, with no link or attachment, can pass every technical check. The last defence is a person who pauses, so the book recommends a one-click report button routed to a monitored mailbox and triage queue, and a blameless culture that thanks over-reporters. Reporting rate should be tracked as a health measure. Whole-staff training is handed off to Book 3 of the series.
KEY POINTS
Key points
- Publish or verify SPF, remove stale senders and move it toward -all.
- Enable DKIM signing for every domain and subdomain that sends mail.
- Publish DMARC at p=none with an rua address and start reading the reports this week.
- Plan the DMARC path from none to quarantine (ramped with pct) to reject, checking reports at each step.
- Turn on impersonation protection, Safe Links and Safe Attachments, and block external auto-forwarding by default.
- Deploy a one-click report button, keep reporting blameless, and require a second signatory for payment-detail changes.
QUESTIONS
Common questions
- What is the difference between SPF, DKIM and DMARC?
- SPF lists in DNS the servers allowed to send mail for your domain. DKIM adds a cryptographic signature to each message so a receiver can confirm it came from your domain and was not altered, and it survives forwarding. DMARC ties both checks to the From address the user actually sees, tells receivers what to do with mail that fails, and sends you reports. The Hardened Stack stresses that you need all three together.
- How do I move DMARC from p=none to p=reject safely?
- Publish p=none with an rua reporting address and read the aggregate reports for several weeks to find every legitimate sender, fixing their SPF and DKIM. When legitimate mail passes reliably, move to p=quarantine, using the pct tag to apply it to a growing share of failing mail. Only after quarantine runs clean should you set p=reject. Jumping straight to reject can silently destroy real invoices and notifications.
- Does DMARC stop phishing from lookalike domains?
- No. DMARC only governs mail that claims to come from your exact domain; a lookalike such as a hyphenated variant is a different domain your record cannot control. Lookalikes are handled by impersonation protection in the mail platform, external-sender warnings, and users who are trained to report suspicious messages.
- Why do technical email controls not stop business email compromise?
- Business email compromise often contains nothing technically wrong: no malware, no bad link, just a convincing request. When it comes from a genuinely compromised internal mailbox it even passes SPF, DKIM and DMARC. The book's answer is process, such as requiring a second signatory for any change to payment details, together with users who feel safe reporting suspicious requests.
- What should I do about external auto-forwarding in Microsoft 365?
- Disable automatic forwarding to external addresses across the tenant by default, allow it only by exception, and alert on any new forwarding rule. Attackers who compromise a mailbox often add a silent forwarding rule so they keep receiving mail after the password is reset. Blocking it removes that tactic, and the alert gives early warning that an account may be compromised.