Prompt
What should I use to detect suspicious logins in a SaaS app?
Latest observation
Oct 23, 2026
OpenAI APIWeb search: off
To detect suspicious logins in a SaaS app, use a combination of risk-based authentication signals, behavior/anomaly detection, and centralized logging/alerting.
Core signals to collect
Track these on every login:
- IP address and ASN / ISP
- Geolocation (country, region, city)
- Device/browser fingerprint or device ID
- User agent
- Login time / time of day
- Failed login attempts
- Velocity: too many logins from one IP or account in a short time
- Session history: new device, new IP, impossible travel, unusual session duration
Useful detections
Flag logins with rules like:
- Login from a new country or far-away location shortly after another login
- Impossible travel (e.g., two distant geos in an unrealistically short time)
- New device + new IP for a sensitive account
- Many failed attempts followed by success
- Login from TOR/VPN/proxy/datacenter IPs
- Login at an unusual time for that user
- Multiple accounts sharing the same suspicious IP/device
- Sudden changes in login patterns or frequency
What to use technically
- Authentication logs in your app backend
- A SIEM / log analytics tool for aggregation and alerting:
- Splunk
- Datadog
- Elastic / ELK
- Microsoft Sentinel
- Google Chronicle
- Fraud/risk detection or identity tooling if you want managed scoring:
- Okta Risk Engine / Identity Threat Protection
- Auth0 Actions + risk signals
- AWS Cognito with custom risk logic
- GeoIP and IP reputation services:
- MaxMind GeoIP
- IPQualityScore
- AbuseIPDB
- GreyNoise
Best practice approach
- Log all auth events
- Enrich events with geo/IP reputation/device info
- Build risk scoring with rules first
- Alert or step-up auth when risk is high
- Continuously tune based on false positives
Good immediate starter set
If you want a practical minimal stack:
- App login logs → Datadog or Elastic
- Geo/IP lookup → MaxMind
- IP reputation → AbuseIPDB or GreyNoise
- Risk rules for:
- new country
- impossible travel
- new device
- VPN/TOR/datacenter IP
- too many failures
If you want, I can also help you design a suspicious login detection schema, risk scoring formula, or sample detection rules for your stack.