Skip to main content

Impossible travel investigation

How an AI SOC works impossible travel alerts: check known egress, the device, and what the session did next before closing as benign or escalating.

Use case
Updated

The short answer

An impossible travel alert says one account signed in from two places too far apart for the time between them. The investigation rules out known explanations first, such as corporate VPN egress that Cotool agents keep in shared global memory, before treating the sign-in as a compromise.[1][8]

The problem

An attacker with stolen credentials signs in as the real user, so the sign-in's location, device, and network are what set them apart.[8]

Microsoft Graph returns Entra ID sign-ins only within the default retention period, so older history has to come from stored logs.[9]

How an AI SOC handles it

  1. Pull both sign-ins

    The agent reads both sign-ins from Okta or Entra ID with IP, device, and conditional access outcome.[2][3]

  2. Check known egress

    IPs are compared with ranges agents already know, such as corporate VPN egress recorded in global memory, and with the user's Zscaler or Cloudflare traffic.[1][3][4]

  3. Check the device

    Intune compliance and Cloudflare Zero Trust device records show whether the second sign-in came from a known device.[2][3]

  4. Look at what came next

    If the sign-in is still unexplained, the agent checks what the session did, such as Drive access or new OAuth grants.[5][10]

  5. Close or escalate

    Explained sign-ins close as Benign with the evidence attached. Unexplained ones are escalated for a person.[6]

Example

Sign-in history for one user in Sentinelkql
SigninLogs
| where TimeGenerated > ago(24h)
| where UserPrincipalName =~ "jdoe@example.com"
| summarize attempts = count(), failures = countif(ResultType != "0"), ips = make_set(IPAddress) by AppDisplayName, Location
| order by failures desc

A KQL query an agent can run through the Log Analytics query API to see where one user signed in over the last day. Replace the user and window for your alert.[7][11]

What people approve

  • If the agent recommends revoking the user's sessions, that action can wait for a person to approve it in the app or in Slack.[1]

How Cotool handles it

  • Cotool keeps what agents learn about your environment, such as which ranges are corporate VPN egress, in shared global memory files that every agent reads and you can inspect.[1]

Integrations used

Sources

  1. AI SOC (Cotool) · accessed November 16, 2026
  2. Microsoft Graph integration (Cotool docs) · accessed November 16, 2026
  3. Changelog (Cotool docs) · accessed November 16, 2026
  4. Zscaler integration (Cotool docs) · accessed November 16, 2026
  5. Detection engineering (Cotool Detect) · accessed November 16, 2026
  6. Alerts (Cotool docs) · accessed November 16, 2026
  7. Microsoft Sentinel integration (Cotool docs) · accessed November 16, 2026
  8. Valid Accounts, T1078 (MITRE ATT&CK) · accessed November 16, 2026
  9. List signIns (Microsoft Graph) · accessed November 16, 2026
  10. Manage OAuth token audit activity (Google Workspace Admin SDK) · accessed November 16, 2026
  11. Azure Monitor Logs query API overview (Microsoft Learn) · accessed November 16, 2026

Related use cases

All use cases

Related reading

Attackers are scaling with tokens

Cotool helps defenders operate at machine speed. See how security teams are scaling Detection & Response.

Book a demo