Security playbook · AI Employee: Sig

Phishing & Suspicious Email Triage

Median analyst time per report < 3 min

The problem

Employees report suspicious emails. Security triages each one manually: is it phishing? malware? just spam? Every triage is 5-10 minutes of security-analyst time. When reports scale — and they always do — either the queue grows or the analyst becomes automatic ("yeah probably fine") and misses the real one.

At a glance
Trigger
Ticket event + email forward
Approvals
Analyst confirmation before mass-remediation
What it does
Writes to your systems
Systems
Email platform · URL sandbox · Threat intel
How it feels in production

An hour-by-hour walkthrough.

9:14am. Priya forwards a suspicious email to [email protected]: fake invoice from what looks like a vendor, urging urgent payment. 9:14am + 30 sec. Sig ingests the reported email. Extracts sender, subject, links, attachments. Sandboxes the links — one goes to a credential-harvesting site hosted on a lookalike domain. Sandboxes the attachment — a Word doc with a malicious macro. 9:15am. Sig classifies: high-confidence phishing (both link + attachment malicious). Immediate actions: quarantine the email from every mailbox that received it (12 people, all forwarded to a suspected-phishing folder), scan mailboxes for related messages (same sender, same subject pattern, 3 more found), and page the security-on-call analyst with the summary. 9:16am. Priya gets a thanks + confirmation: "Confirmed phishing. Quarantined from 12 mailboxes. Nice catch." 9:20am. Security-on-call analyst opens Slack. Reviews Sig's classification. Confirms. Marks resolved. Optionally: publishes a security-awareness note ("the phishing you might have seen this morning was quarantined; here's what it looked like") — Sig drafts it.
How it works

Step by step.

  1. 01

    Ingest reported email

    Email forwards to [email protected]. Sig extracts headers, sender, subject, body, links, attachments. All processed in isolated sandbox — nothing executes in your environment.

    Email gateway · Slack integration
  2. 02

    Sandbox links + attachments

    Every URL clicked in an isolated browser (URL sandbox); every attachment opened in an isolated VM (attachment sandbox). Malicious behaviour observed: credential harvesting, malware download, exfiltration.

    URL sandbox · Attachment sandbox · Threat intel
  3. 03

    Classify

    Classification model + threat-intel enrichment: phishing (harvest credentials), malware (executable payload), BEC (business email compromise / wire fraud), spam (annoying but harmless), legitimate (falsely reported).

    LLM · Threat intel APIs
  4. 04

    If malicious: quarantine + hunt

    Quarantine the reported email from every mailbox that received it. Hunt for related messages (same sender, similar subject, similar template). Auto-quarantine the related ones. Notify affected recipients.

    Email platform · Slack
  5. 05

    Analyst review

    Security analyst reviews Sig's classification + actions taken. Rare misclassifications get corrected + fed back for future accuracy. Common patterns become detection rules.

    Slack · Security dashboard
Systems and wiring

What you connect to make this run.

Google Workspace · Microsoft 365

read+write

Email API with read + quarantine scopes. Read reported emails; quarantine originals + related messages from affected mailboxes.

URL sandbox (urlscan.io · VirusTotal · custom)

read

External sandbox APIs. URLs submitted, verdict returned within seconds. No traffic ever touches your production environment.

Attachment sandbox (Joe Sandbox · Any.Run · custom)

read

Attachments detonated in isolated VMs. Behavioural report returned; findings correlated with threat intel.

Threat intel (VirusTotal · abuse.ch · commercial feeds)

read

IOCs from the email checked against known-bad indicators. Enrichment improves classification confidence.

What changes

Before and after, honestly.

Median analyst time per report
Before
10-15 minutes across sandbox + classification + action
After
Under 3 minutes (review Sig's work)
Time from report to quarantine
Before
30-90 minutes
After
Under 2 minutes for high-confidence cases
% of malicious emails discovered proactively (before more employees clicked)
Before
40-60% (queue backlog delays action)
After
85-95% (mailbox hunt runs on every classified phish)
Employee experience reporting phishing
Before
"I forwarded it, no idea if anyone saw"
After
Confirmation within minutes; visible action taken
Frequently asked

Answers about this playbook.

What if the phishing is targeted (spear-phishing, not mass)?

Same flow. Sig quarantines the reported email + hunts for related; even single-recipient spear-phishing gets analyst attention within minutes. Detection is not sensitive to campaign size.

What about business-email-compromise (BEC) — email that looks like it's from the CEO?

Classification includes BEC detection: sender-domain analysis (lookalike domains flagged), tone analysis (urgency + wire-transfer request), and cross-check against actual sender via DKIM. High-confidence BEC gets escalated immediately.

Can we integrate with our SEG (secure email gateway)?

Yes — SEG detections feed Sig as high-confidence signals. Sig complements the SEG by handling the human-reported cases the SEG missed. Analysts see both feeds unified.

What if the reported email turns out to be legitimate?

Sig classifies as legitimate + closes the report with a note to the reporter. Feedback loop: consistent false-positive senders get whitelisted; analyst can retrain the classifier on the mis-classification.

Does this violate email privacy?

Only quarantines emails that were reported by recipients OR match a high-confidence threat pattern. Never reads employee mailboxes at will. Full audit trail of every quarantine action.

See it run on your data.

Free plan, no credit card. Connect the systems this playbook needs and run it against a past event first.