Phishing7 Oct 2026· 6 min read
Most email security demos use cartoonish lures — misspelled banks, Nigerian
princes — and every filter catches them. Real attacks don't look like that.
The messages that slip through are short, plausible, and borrow trust from
context the filter can't see.
The patterns that evade content filters
- Thread hijacking. A reply injected into a genuine, ongoing
conversation inherits the trust of the original thread.
- Look-alike domains.
rnicrosoft.com or an IDN homoglyph
reads as the real thing at a glance and passes SPF/DKIM for the attacker's
own domain.
- No payload in the body. The malicious step is a benign-looking link
to a credential page hosted on a reputable platform, so there's nothing
obviously bad to scan.
The lesson for anyone evaluating a detection product: test it against the
boring, well-crafted messages, not the cartoons. A tool that scores 100% on
obvious spam tells you nothing about the mail that actually causes incidents.
Email auth2 Oct 2026· 5 min read
These three records decide whether your mail is trusted — and whether someone
else can send as you. They're often half-configured, which quietly hurts
deliverability and leaves the door open to spoofing.
What each one actually does
- SPF lists which servers are allowed to send mail for your domain.
Receivers check the sending server against that list.
- DKIM adds a cryptographic signature to each message, so the receiver
can verify it wasn't altered and really came from your domain.
- DMARC ties the two together and tells receivers what to do when a
message fails — monitor, quarantine, or reject — and where to send reports.
A common mistake is stopping at p=none forever. It's the right
starting point for monitoring, but leaving it there means you see spoofing in
the reports without ever blocking it.
Identity24 Sep 2026· 7 min read
Federated login removes passwords, but it moves the risk into how your app
reads the identity token. Two mistakes show up again and again in testing.
Mistake 1: trusting the email address
Email is not a stable identifier. Providers let it change, and the same address
can belong to a personal account one day and an organisation account the next.
Link accounts by the provider's immutable subject identifier, not the address.
Mistake 2: ignoring the account type
A token from a managed organisation account carries different claims from a
personal one. Apps that don't distinguish them can accidentally auto-join users
into the wrong workspace. Always check the claim that identifies the tenant,
not just the domain of the email.
Testing16 Sep 2026· 5 min read
"It caught the phishing email" is not a metric. To compare tools, or to know
whether a release got better or worse, you need numbers against a known set of
labelled messages.
The two that matter
- Recall — of all the malicious messages, how many did it catch?
Low recall means threats get through.
- Precision — of everything it flagged, how much was actually
malicious? Low precision means users drown in false alarms and start
ignoring the tool.
These trade off against each other, so a single accuracy figure hides what's
really happening. We run every candidate against the same labelled corpus and
report both, release over release, so improvements are provable rather than
anecdotal.
Safety9 Sep 2026· 4 min read
You can validate that a scanner reacts to a malicious attachment without ever
handling a live sample. The EICAR test file is a harmless, industry-standard
string every antivirus engine recognises as a trigger.
It lets you prove the plumbing works — that an attachment is scanned, flagged,
and handled correctly — with zero risk to the test environment or the people
running it. Real samples belong in isolated, purpose-built labs, never in a
shared mailbox or a CI pipeline.