Okta System Log
Okta System Log event types, indexed by what each one tells an investigator.
Compromise an identity provider and every application behind it follows. Okta's System Log records source ASN, proxy status, device fingerprint and its own threat assessment on ordinary sign-ins. This is the subset of several hundred event types that changes what you do next.
Before you read the table
- Pivot on externalSessionId, not the username. Following the user merges the attacker's activity with the real user's.
- A push is only sent after the password succeeds. A run of push verifications means the password is already known.
27 rows.
| Event | Meaning | Investigative use |
|---|---|---|
| user.session.start | A user session was established. | The anchor event for any Okta timeline. Read `outcome.result` (SUCCESS / FAILURE), `outcome.reason`, and the `client` block. Okta records the IP, user agent, and its own geolocation and device fingerprint on every one. `authenticationContext.externalSessionId` is what stitches every later event in the same session together. |
| user.session.start (FAILURE, INVALID_CREDENTIALS) | A sign-in failed because the password was wrong. | The spray and brute-force signal. As everywhere else, breadth across many distinct users from one source is a spray; depth against one user is brute force. Okta's `securityContext` block adds the source ASN and whether the address is a known proxy, which is more context than most identity logs give you for free. |
| user.authentication.auth_via_IDP | Authentication was delegated to an external identity provider. | The seam between Okta and an upstream IdP. If an inbound federation was added recently, this is where sessions from it start appearing, and a newly-added IdP is a well-known Okta backdoor technique. |
| user.authentication.sso | A user was signed in to a downstream application via SSO. | The record of what the session actually reached. High volume and unremarkable individually; decisive when you need to know which applications a confirmed-compromised session touched. |
| user.session.access_admin_app | A user opened the Okta administrator console. | Everyone who administers your identity provider produces this, and nobody else should. An account that has never opened the admin console suddenly doing so is one of the cleanest single signals in the whole log. |
| system.push.send_factor_verify_push | A push verification was sent to the user's device. | Counting these is how you detect push-bombing. A rapid run against one user means the attacker already holds a valid password. The push only happens after the first factor succeeds. The event to look for next is the first successful verification after the run. |
| user.authentication.auth_via_mfa | An MFA factor was verified (or failed). | Read `outcome.result` alongside the factor type. A sudden change in which factor an account uses, from a hardware key to SMS say, is often the attacker steering the user toward the weakest available option. |
| user.mfa.factor.activate | A new MFA factor was enrolled for a user. | An attacker enrolling their own authenticator converts a stolen password into access that survives the reset you are about to perform. Any enrollment during an incident window, or from an address the user has never signed in from, should be treated as hostile until the user confirms it. |
| user.mfa.factor.deactivate | An MFA factor was removed. | Frequently paired with an immediate re-enrollment of an attacker-held device. On its own it weakens the account; in a pair it is a takeover. |
| user.mfa.attempt_bypass | An attempt was made to bypass MFA. | Rarely legitimate and explicitly named. Treat any occurrence as an incident until proven otherwise. |
| user.account.reset_password | A user's password was reset. | Compare the actor against the target. A self-service reset immediately after a suspicious sign-in is the attacker locking the legitimate owner out; an admin-initiated reset should have a help-desk ticket behind it. |
| user.account.lock | An account was locked after repeated failures. | Confirms the failure volume crossed Okta's threshold. A cluster of locks across unrelated users at the same moment is an attack, not a coincidence, and it is often the first thing the service desk notices. |
| user.lifecycle.create | A new user was created in the directory. | In an environment provisioned from an HR system, a user created directly in Okta bypassed that process entirely. Check whether the creating actor normally provisions users at all. |
| user.lifecycle.deactivate | A user was deactivated. | Ordinary in a leaver workflow; destructive when it is unexpected, and a useful cover for hiding an account that was used. |
| group.user_membership.add | A user was added to a group. | Privilege escalation by indirection. Okta groups typically drive both application assignment and admin roles, so the resulting permissions are one lookup away from this log line. Resolve what the group actually grants before dismissing it. |
| policy.lifecycle.update | A policy was modified. | Easy to miss and wide-reaching. Weakening a sign-on or MFA-enrollment policy, by adding a network zone exception or dropping a factor requirement, degrades security for everyone it covers while producing exactly one log line. Read what changed, not that something did. |
| policy.lifecycle.deactivate | A policy was deactivated. | Same reach as modifying one, with nothing left to inspect afterwards. Reconstruct from the policy's own history. |
| policy.evaluate_sign_on | A sign-on policy was evaluated for an authentication attempt. | Very high volume, and the record of which rule matched and what it decided. Its value is explaining why a sign-in that should have been challenged was not. |
| zone.update | A network zone was modified. | Zones drive policy exceptions, so adding an address to a trusted zone can exempt it from MFA without touching any policy object. A change here plus a later sign-in from the added range is a complete bypass in two events. |
| system.api_token.create | An Okta API token was created. | One of the most durable persistence mechanisms available in Okta. An API token carries the privileges of the admin who created it, is not subject to MFA, does not expire on a password change, and is rarely inventoried. Every creation should map to a named integration. |
| system.api_token.revoke | An API token was revoked. | Useful for confirming your own containment took effect, and for spotting an attacker revoking a token to cover their tracks. |
| app.oauth2.client.credentials | An OAuth 2.0 client-credentials grant was issued. | A workload identity authenticating with no user and no MFA. Legitimate for integrations, and an attractive target precisely because it is exempt from the human-facing controls. |
| application.lifecycle.update | An application configuration changed. | Read the diff. Changing an app's sign-on mode, its SAML assertion attributes, or its redirect URIs can redirect authentication or leak tokens, and none of that is visible from the event name. |
| application.user_membership.add | A user was assigned to an application. | How an attacker reaches a downstream system they were not previously entitled to. Read alongside user.authentication.sso to see whether the assignment was then used. |
| user.session.impersonation.initiate | An administrator began impersonating a user. | Everything done during impersonation happens as the target user, which makes this the single most important event to review by hand. It is a supported feature that grants complete, attributable-to-someone-else access. Pair every initiate with its matching `user.session.impersonation.end` and account for the gap between them. |
| user.session.impersonation.end | An impersonation session ended. | The closing bracket. An initiate with no matching end means the session was still open when the log window closed. |
| security.threat.detected | Okta ThreatInsight flagged the request as malicious. | Okta's own assessment, drawn from activity across its whole customer base, so it sees source addresses attacking other tenants before they reach yours. Whether it blocked or merely logged depends on the tenant's ThreatInsight configuration; check which, because "detected" does not imply "prevented". |