Obsidian Email SecurityDocumentation Sign in

Test a message

Monitor › Test a message runs the complete scan on a message you upload and shows exactly what Obsidian Email Security would do - without delivering, storing or learning anything.

Getting an .eml file

  • Outlook (classic desktop): it saves .msg files, which are not .eml and cannot be scanned. Open the same message in Outlook on the web instead (below).
  • Outlook on the web / new Outlook: message › … › Save as (downloads .eml).
  • Gmail: message › ⋮ › Show original › Download original.
  • Thunderbird: File › Save As › File.
  • This console: any message page › Download .eml.

Fields

Message file
The .eml (up to 60 MB).
Recipient
A protected address. Picks the organization and therefore its domains, VIPs, policy overrides, lists and Bayes corpus. Leave empty to test with the default policy only.
Envelope sender
The SMTP MAIL FROM (used for SPF). Take it from the message's Return-Path header.
Connecting IP
The IP that delivered it (from the first external Received header). Enables SPF and Spamhaus ZEN; without it those checks are skipped.

Reading the result

  • The header shows the verdict, the action that would be taken, the score and the reason.
  • The AI runs inline here (if it would be asked) so you see its opinion - allow up to the configured AI timeout.
  • Stages shows each check's raw data, exactly as on a real message page.
  • Headers that would be added shows the X-Obsidian headers and Authentication-Results the recipient would receive.
Authentication depends on context. A message re-sent from a file may fail SPF/DKIM that passed originally (DKIM breaks if the file was modified, SPF depends on the IP you enter). Compare with the original Authentication-Results header.

AI test

Monitor › AI test (superadmins) checks the AI model on its own, without DNS lookups or the rule engine:

  • Run AI self-test sends five built-in messages whose right answer is known - credential phishing, an executive gift-card request (BEC), an invoice with a disk-image attachment, a plain note from a friend with no authentication, and a newsletter - and shows the model's verdict, risk and time for each. pass* means a neighbouring verdict that still protects the recipient. Run it after changing the model or its settings: a model that calls the friend's note malware will quarantine real mail.
  • Ask the AI takes a message you write or paste (From, subject, body, the SPF/DKIM/DMARC results the gateway would have found, and an optional attachment name) and shows the verdict, risk, summary and evidence.
  • Both show what the model saw - the exact text it received, in the same format as live mail.
  • They run on the AI worker behind real mail reviews, so on a busy gateway an answer can take longer. Nothing is delivered or written to the message log.
← Quarantine Policies →