~/blog/phishing-mail-in-four-decisionsPOSTED
by Shady Nathan Tawfik · November 16, 2026 · 7 min

A Phishing Mail Is Four Decisions

No malware, no zero-day, no lookalike domain: a phishing mail is four decisions — who appears to send, how it travels, what the link says, and what it asks for. Measured on a domain I control, with the header lines that decide the outcome.

Interactive[blog-2.0]

TL;DR: A phishing mail needs no exploit. Four decisions do the work: a believable sender, a transport path that survives the MX, a link that reads like the brand, and one small ask under time pressure. In a controlled test on a domain I control, two of three mails reached the inbox. The one that failed was stopped by the domain policy, not by the content filter. The whole build took an afternoon and about forty lines of script.

The scenario

I test mail chains for a living, so I built three phishing mails against a domain I own and a mailbox I can read back. One recipient, three variants, one message each. No attachments, no macros, no credential form, no third-party brand.

The result is not the interesting part. What interested me was how little there was to write.

Fourteen lines of text produce a mail that looks like a business notification. Four decisions decide whether it lands in the inbox or in the spam folder — and only one of those decisions is about the content.

This is a pattern essay. There is no client, no country, no hostname and no real domain in it. The mechanics are unchanged; the target is not recoverable.

Decision one: who appears to send

Every mail carries two senders, and most people only ever see one of them.

~/code/text text
envelope      MAIL FROM:<bounce@relay.tld>     ← the server talks to this
header        From: service@example.tld         ← the human sees this
display name  "Portal Benachrichtigung"         ← 100 % free text

Authentication checks the envelope domain and the domain inside From. The display name is not part of any check — it is a text field. That gives three ways to look like a brand:

  1. The brand's own domain, through the brand's own mail provider. Nothing is faked. The message is genuinely signed with the domain's key.
  2. Your own domain, with the brand in the display name. The domain is honest, the name is not.
  3. The brand's domain in From, with no aligned signature. Classic spoofing — and the only one of the three that a DMARC policy can stop.

The first two need no vulnerability at all.

Decision two: how it travels

There are three roads to a mailbox, and they behave very differently.

Road A — the brand's own provider. If you can use the same sending service the brand uses, the signature is real: dkim=pass with the brand's domain. This is the strongest path and the one defenders underestimate, because it looks like a legitimate message. Technically, it is one.

Road B — straight to the recipient's MX. This is the version people imagine when they hear "spoofing". It is also the version that fails first. Two attempts, two refusals:

~/code/text text
550-5.7.1 … does not meet IPv6 sending guidelines regarding PTR records and authentication
550-5.7.1 The IP you're using to send mail is not authorized to send email directly to our servers

Both happen at DATA — before DMARC is even evaluated. A random server on the internet cannot deliver directly, full stop.

Road C — a rented relay. A normal hosting server accepts the message, rewrites Return-Path to its own domain, adds an ARC chain and forwards it. SPF passes for the relay, arc=pass, and the only thing left between the message and the inbox is DMARC — which fails, because nothing is aligned with the domain in From.

In my test, road C ended in spam. Not in the bin. In spam.

No lookalike domain. No homograph. No shortener, no open redirect.

~/code/text text
https://example.tld@my-host.tld/confirm

Everything before the @ is a username; the real host follows it. A browser goes to my-host.tld, a reader sees the brand at the start, and a link scanner sees a clean, reachable page.

The delivered HTML contained exactly that link in all three variants — unchanged, no rewrite, no warning interstitial, no block. The only cost of this decision is a host you already own.

Decision four: what it asks for

This is the only part a content filter can judge, so it stays small:

  • one click,
  • twenty-four hours,
  • no attachment,
  • no password field in the mail,
  • no name instead of a wrong name ("Hallo," beats "Sehr geehrter Herr Müller" when Müller does not exist).

The mail carries a link. What happens after the click is a separate decision — and in this test the landing page had no form at all, because the goal was to measure the chain, not to collect anything.

TL;DR: A phishing mail needs no exploit. Four decisions do the work: a believable sender, a transport path that survives the MX, a link that reads like the brand, and one small ask under time pressure. In a controlled test on a domain I control, two of three mails reached the inbox - the one that failed was stopped by the domain policy, not by the content filter. The whole build took an afternoon and about forty lines of script.

Findings

DecisionCheap versionReal costWhat the defender sees
Senderbrand domain via its own provideraccess to that providerdmarc=pass, correct domain, nothing to flag
Senderown domain + brand display nameone domain you owndmarc=pass, wrong domain in From
Senderforged domaina relay that accepts foreign sendersdmarc=fail
Transportdirect to MXa server550 at DATA (PTR, authorization)
Transportrented relayone host, one scriptarc=pass, spf=pass for the relay, dmarc=fail
Linkuserinfo prefixnothingURL starts with the brand, host is elsewhere
Askone click, 24 hoursa pagenothing technical at all

The chain

Choose senderBrand domainOwn domain + brand nameForged domainOwn providerRented relayMailboxLink with brand prefixLanding pageCredential harvestPayment change

What the measurements showed

Three variants, three outcomes, and only one of them was a decision by the receiving side:

  • Authenticated brand mail → inbox. The filter had no reason to act. The message was technically flawless.
  • Brand display name, foreign domain → inbox. No lookalike warning, no reply-to mismatch flag.
  • Forged sender via relay → spam. dmarc=fail with dis=QUARANTINE. Quarantined, not rejected — the message still exists in the mailbox.
  • Direct spoof → refused at SMTP. The one path that never became a message.

Read that as a defender and the conclusion is uncomfortable: the layer that "works" is authentication, and it only works against the laziest variant. The content layer changed nothing in any of the three.

Detection: the thirty seconds that matter

For a person reading the mail:

  1. Read the domain after the @ in the link. The text before it is decoration.
  2. Open the header and find Authentication-Results. dmarc=pass means "not forged" — it does not mean "trustworthy". A perfectly authenticated message can still be a lie.
  3. Compare the From domain with the Return-Path domain. A mismatch is normal for bulk mail and worth a second look everywhere else.
  4. Treat the display name as decoration. It is the cheapest thing to fake and the first thing a reader trusts.

For the domain owner, the four records that turn decisions one to three from "depends" into "no":

~/code/text text
spf      v=spf1 … -all
dmarc    v=DMARC1; p=reject; sp=reject; rua=mailto:…
tls-rpt  v=TLSRPTv1; rua=mailto:…
caa      0 issue "…"

And the honest footnote: none of these covers decision one, case two. Display-name spoofing is not a DNS problem.

What each decision costs the defender

Phishing is cheap because the attacker's costs are near zero and the defender's are not:

Attacker paysDefender pays
a domain or an existing accountmailbox filtering, user training, incident response
one host for a relaymail authentication, monitoring, DMARC reporting
an afternoona process that survives a bad afternoon
nothing for the display namebrand monitoring and lookalike detection

That asymmetry is the actual finding. It is not that a phishing mail is clever. It is that building one is cheaper than defending against one — and the gap is widest exactly where nobody looks: the display name, the URL prefix, and the mail a company sends to itself.

Closing

A phishing mail is not a hack. It is a message that answers four questions convincingly enough that nobody asks a fifth.

The good news is that three of the four leave traces in the headers, and the fourth leaves a trace in the URL. The bad news is that those traces are only visible to someone who knows where to look — and almost nobody does, because the mail looks exactly like the ones they receive every day.

Once an attacker gains access, they can perform the following actions:

Harvest credentials with a landing page that looks like the real login - the mail itself never contains a password field, so nothing technical stands out.
Send a payment or bank-detail change as a routine notification, from a domain the recipient already trusts.
Wear the brand in the display name without registering a lookalike domain - the cheapest form of brand abuse there is.
Use a rented relay so the message comes from a reputable envelope domain and only the DMARC policy decides its fate.
Put the brand at the start of the URL with a userinfo prefix - readers check the beginning, not the host.
Shady Nathan Tawfik

Founder & Cyber IT Security Engineer in Bielefeld. Offensive security, audits and full-stack builds. More at

For educational purposes only. Every reproducible command targets my own lab environment (localhost), never a live system. © 2026