callback_phishing_nlu_body_or_attachments.yml

name: "Callback Phishing in body or attachment (untrusted sender)"
description: |
  Detects callback scams by analyzing text within images of receipts or invoices from untrusted senders.
type: "rule"
severity: "medium"
source: |
  type.inbound
  and length(attachments) < 5
  and (
    any(attachments,
        (.file_type in $file_types_images or .file_type == "pdf")
        and any(file.explode(.),
  
                // exclude images taken with mobile cameras and screenshots from android
                not any(.scan.exiftool.fields,
                        .key == "Model"
                        or (
                          .key == "Software"
                          and strings.starts_with(.value, "Android")
                        )
                        or (.key == "UserComment" and .value == "Screenshot")
                )
                and any(ml.nlu_classifier(.scan.ocr.raw).intents,
                        .name == "callback_scam"
                        and .confidence in ("medium", "high")
                )
        )
    )
    or any(ml.nlu_classifier(body.current_thread.text).intents,
           .name in ("callback_scam")
           and .confidence in ("medium", "high")
           and length(body.current_thread.text) < 1500
    )
  )
  and not (
    any(headers.domains, .domain == "smtp-out.gcp.bigcommerce.net")
    and strings.icontains(body.html.raw, "bigcommerce.com")
  )
  and (
    not profile.by_sender().solicited
    or (
      profile.by_sender().any_messages_malicious_or_spam
      and not profile.by_sender().any_false_positives
    )
  )
  
  // negate highly trusted sender domains unless they fail DMARC authentication
  and (
    (
      sender.email.domain.root_domain in $high_trust_sender_root_domains
      and not headers.auth_summary.dmarc.pass
    )
    or sender.email.domain.root_domain not in $high_trust_sender_root_domains
  )

attack_types:
  - "Callback Phishing"
tactics_and_techniques:
  - "Out of band pivot"
  - "Social engineering"
detection_methods:
  - "Content analysis"
  - "File analysis"
  - "Optical Character Recognition"
  - "Natural Language Understanding"
  - "Sender analysis"
id: "b93c6f94-c9a3-587a-8eb5-6856754f8222"