Description

Detects inbound messages containing DOCX attachments with compensation or benefit-related themes that include QR codes and suspicious indicators. The rule identifies files with reward/benefit language in filenames, compensation-related content in document metadata, and QR codes that may redirect to credential theft pages. It uses natural language processing to detect credential theft intent and suspicious topics like benefit enrollment or financial communications.

References

No references.

Sublime Security
Created May 29th, 2026 • Last updated May 29th, 2026
Source
type.inbound
and (
  length(filter(attachments, .file_type == "docx")) >= 1
  and (
    // short or null message body
    (
      length(body.current_thread.text) < 500 or body.current_thread.text is null
    )
    // ignore disclaimers in body length calculation
    or (
      any(map(filter(ml.nlu_classifier(body.current_thread.text).entities,
                     .name == "disclaimer"
              ),
              .text
          ),
          (length(body.current_thread.text) - length(.)) < 500
      )
    )
  )
)
and (
  // attached DOCX contains a compensation review themed lure with a QR code and suspicious indicators
  any(filter(attachments, .file_type == "docx"),
      // add conditions for DOCX attachment
      (
        regex.icontains(.file_name,
                        '(?:salary|pay(?:roll)|bonus|comp(?:ensation|liance|\b)|remuneration|disbursement|incentive|merit|vesting|employee.*(?:reward|benefit)s?)'
        )
        // recipient email SLD in filename
        or any(recipients.to,
               strings.icontains(..file_name, .email.domain.sld)
               and .email.domain.valid
        )
        or regex.icontains(beta.parse_exif(.).title,
                           '(?:salary|pay(?:roll)|bonus|comp(?:ensation|liance|\b)|remuneration|disbursement|incentive|merit|vesting|employee.*(?:reward|benefit)s?)'
        )
      )
      // add conditions for text and any QR code within the DOCX attachment
      and (
        // conditions for QR code via text
        any(file.explode(.),
            any([.scan.strings.raw, .scan.ocr.raw],
                regex.icontains(., 'scan|camera|review and sign')
                and regex.icontains(., '\bQR\b|Q\.R\.|barcode')
            )
            or (
              .scan.qr.type == "url"
              and .scan.qr.url.url is not null
              and any(recipients.to,
                      .email.domain.valid
                      and (
                        strings.icontains(..scan.qr.url.url, .email.email)
                        or any(strings.scan_base64(..scan.qr.url.url,
                                                   format="url"
                               ),
                               strings.icontains(., ..email.email)
                        )
                      )
              )
            )
        )
        or any(file.explode(.),
               .scan.qr.type == "url" and .scan.qr.url.domain.valid
        )
      )
      // conditions for text
      and any(file.explode(.),
              // review/change terms in file content
              any([.scan.strings.raw, .scan.ocr.raw, .scan.exiftool.title],
                  (
                    regex.icontains(.,
                                    '\b(?:Remuneration Overview|Updated Compensation (?:Summary|Schedule|Details)|Access Your Statements?|Staff Performance Appraisal|Compensation Adjustment|performance appraisal|Appraisal Overview|appraisal and compensation|salary (?:increment|deduction))\b'
                    )
                  )
              )
              or (
                // recipient local_part in attachment body
                any(recipients.to,
                    strings.contains(..scan.ocr.raw, .email.local_part)
                )
                and (
                  // NLU cred_theft disposition
                  any(ml.nlu_classifier(.scan.ocr.raw).intents,
                      .name == "cred_theft" and .confidence != "low"
                  )
                  // suspicious topics
                  and any(ml.nlu_classifier(.scan.ocr.raw).topics,
                          .name in (
                            "Benefit Enrollment",
                            "Financial Communications"
                          )
                          and .confidence != "low"
                  )
                )
              )
      )
  )
)

// negate highly trusted sender domains unless they fail DMARC authentication
and not (
  sender.email.domain.root_domain in $high_trust_sender_root_domains
  and coalesce(headers.auth_summary.dmarc.pass, false)
) 
MQL Rule Console
DocsLearning Labs

Playground

Test against your own EMLs or sample data.

Share

Post about this on your socials.

Get Started. Today.

Managed or self-managed. No MX changes.

Deploy and integrate a free Sublime instance in minutes.
Get Started