Problem
Slack gives Enterprise customers audit logs with anomaly flags, for example irregular logins, malware uploads and unexpected data transfers. Someone still has to read the flags and act, and many customers lack the staff or tools for it. The delay between detection and response, from hours to days, lets an attacker finish an exfiltration. The platform handles billions of events a day, so the response has to be automatic. It also has to fit organizations with very different usage patterns.
Illustrative example: a user's sessions start scraping data through a non-Slack-native automation tool; the detection fires, all of that user's sessions end, and the user, the Org Primary Owner and Security Admins are told.
Solution in brief
Detect high-confidence anomalies with rules and per-organization thresholds, check each detection against the customer's settings and the account's recent history, then terminate all sessions of the acting user. Every response is audit-logged, and notifications are configurable and deduplicated.
Questions this case answers
- How to close the gap between detecting suspicious account behavior and acting on it when most tenants have no security staff to review alerts
- How to calibrate detection thresholds per tenant when the normal level of an activity differs between tenants and a single global threshold would produce false positives
- How to turn anomaly alerts into an automatic response so that abuse is stopped in minutes instead of waiting for an analyst?
- How to set anomaly thresholds when what is normal differs a lot between customer organizations?
Tools and methods
- Rule-based detections — six threat types covering the most common platform risks.
- Per-organization dynamic thresholds — computed from each organization's historical data.
- Decision framework — checks the customer's configuration and the account's history before acting.
- Asynchronous response job — terminates sessions, writes the audit log and routes notifications.
Google or a one-time email code. No password.