Audit
What is recorded for every credential read, check-out, approval, rotation and test, on the platform, in the runner's local audit file and in your PAM's own audit, and how to filter and export the tenant-wide log.
Audit
Every read of a credential, by a scan or by an access test, and every change to a credential with a vault policy, is recorded in three places. None of them ever holds the value.
| Where | What it holds | Who sees it |
|---|---|---|
| Platform audit trail | Action, time, who (a person, an AI agent, the runner's signed-in identity or the scanner), entry, provider, scan, runner, target, scan type, lease, request, version, success or error code | People who can see the entry, under Who has seen this, and everyone with the auditor or admin role in the Audit log tab |
| Runner's local audit file | Everything above plus the full reference, the configured provider name, what the credential was for, the duration and the number of tries | Whoever can read the file on the runner machine; collect it into your SIEM |
| Your PAM's own audit | The retrieval by Safeguard's identity, with the reason text where the PAM records one | Your PAM administrators |
Platform audit trail
Two views show the same rows. Open an entry and choose Who has seen this for that entry's history, or open the Audit log tab of the Credentials Vault page for the whole organization. Each row shows:
| Field | Meaning |
|---|---|
| Time | When it happened |
| Action | See the table of actions below |
| Outcome | OK, ERROR, DENIED or NOT_FOUND; for a failed read the note names the code, for example fetch failed, code FORBIDDEN |
| Actor | The signed-in user or service that acted. For a runner, the identity it signed in with |
| On | The runner name (SAFEGUARD_RUNNER_NAME, default the host name) |
| Scan | The scan or job the read was for |
| Via | The PAM provider type, for a PAM entry |
| Target | What the scan ran against: the URL, host or repository host |
| Scan type | dast, api, infra_host, source or scm |
| Lease, Request, Version | The lease the read belonged to, the approval request it used, and the version of the value that was read |
| Duration | How long the read took, in milliseconds |
| Message | A short cleaned note of at most 500 characters. It never holds a value |
Actions
| Action | Meaning |
|---|---|
CREATE, UPDATE, DELETE | A credential was added, changed or removed |
REVEAL | A person or runner read a stored value |
REVEAL_DENIED | A reveal was refused: write-only, no REVEAL grant, or no recent sign-in |
RESOLVE | Safeguard's cloud scanner read a stored value of a credential with no policy |
FETCH | A scan read a credential (the runner or CLI reports this) |
TEST | An access test |
CHECKOUT | A credential with a policy was checked out, with the lease |
CHECKIN | The scan checked the credential back in |
LEASE_RENEW, LEASE_EXPIRE, LEASE_REVOKE | A lease was renewed, ran out, or was revoked by someone with MANAGE |
REQUEST | A scan or person asked for approval |
APPROVE, DENY | An approver decided, with the comment |
VERSION_CREATE | A new version was stored, by a person or by a rotation |
ROTATE | A rotation was requested, claimed by a runner, staged, completed or failed, with its code |
POLICY_UPDATE, GRANT_UPDATE | The policy or the grants changed |
EXPORT | Someone exported the audit log |
KEY_ROTATE | The organization's encryption key was rotated |
A scan that uses a direct pam:// or vault:// reference (not a vault entry) is recorded in your organization's audit trail without an entry. To see such reads under an entry, use sgvault:// entries in scan targets.
The tenant-wide audit log
Open Settings, Credentials Vault, Audit log. The table lists Time, Who, Action, Credential, Outcome and Detail, newest first, and Load more pages through older rows. Filter by From and To, Credential, Action, Who and Scan.
Who sees what:
- People with the
vault-adminorvault-auditorrole see every row of the organization. - Everyone else sees the rows of credentials they can view.
Export
Choose Export CSV or Export JSONL. The export uses the filters you set and is itself recorded in the log as an EXPORT row. It holds up to 50,000 rows; if more match, narrow the dates and export again (the response says when it was cut short). Fields, in order:
time, actorKind, actorSubject, actorEmail, action, credentialId, credentialName, credentialHandle, provider, ref, version, leaseId, requestId, scanRef, scanType, target, runnerName, outcome, durationMs, message, note, id
CSV cells that start with a formula character are neutralized so a spreadsheet does not run them. From a terminal, safeguard vault audit --export csv or --export jsonl does the same. See the CLI reference.
The runner's local audit file
Every read on a runner or CLI machine, by a scan, a safeguard secrets test or an access test, is appended as one JSON line to vault-audit.jsonl next to the provider file (by default ~/.safeguard/vault-audit.jsonl). It is written even when the platform cannot be reached.
- Created with mode 600 in a folder with mode 700, appended to, never rewritten.
- Rotated to
vault-audit.jsonl.1when it passes 10 MB (the previous.1is replaced). Collect it with your log shipper before then. audit_file: /var/log/safeguard/vault-audit.jsonlin the provider file, orSAFEGUARD_VAULT_AUDIT_FILE, moves it;offturns it off.
One line, formatted here for reading:
{
"time": "2026-11-21T09:14:03.512Z",
"action": "FETCH",
"ok": false,
"code": "FORBIDDEN",
"reference": "sgvault://dast-login",
"entry": "dast-login",
"credentialId": 812,
"provider": "arcon",
"providerName": "arcon",
"target": "the DAST login password",
"scan": "3f9a2c1e-6b1d-4c55-9a0e-7e2b5d1c8f40",
"runner": "dc1-runner",
"durationMs": 412,
"attempts": 1,
"pid": 48211
}| Field | Meaning |
|---|---|
time | UTC, RFC 3339 |
action | FETCH for a scan, TEST for an access test or safeguard vault test |
ok, code | Result; code is one of the error codes |
reference | The reference as written in the scan (sgvault://..., pam://...) |
entry, credentialId | The vault entry, for an sgvault:// read |
provider, providerName | Provider type and its name in the provider file |
target | What the credential was for, for example the DAST login password or the repository access token (--git-token) |
scan, runner | The scan or job id, and the runner's name |
durationMs, attempts | Time the read took, and how many requests (more than 1 after a retry) |
refresh | true when the scan read the credential again after the target refused it |
scanTarget, scanType | What the scan ran against and its type, when the scan said so |
leaseId, requestId, version | For a credential with a policy: the lease, the approval request used, the version read |
message | A cleaned note about a failure, at most 500 characters, never a value |
pid | The runner process |
The same fields also go to the runner's ordinary log as the line secret reference resolved.
Your PAM's own audit
- CyberArk CCP records the
Reasonthe runner sends with each request: by defaultSafeguard scan <scan> on <runner>, or yourreasonsetting with{{scan_ref}},{{runner}}and{{target}}filled in. - BeyondTrust records the request reason (
reasonsetting) and the check-in at scan end. - ARCON PAM and the REST connector record what their API records. Add
{{reason}}or{{scan_ref}}tofetch_body_templateorfetch_querywhere the API takes a reason or comment field. - HashiCorp Vault records the read in its audit device under the runner's token or AppRole.
To trace one read across all three, match the scan id and the runner name: they are in the platform row, in the local file, and in the reason text the PAM records.
Delivery to the platform
The runner posts each record to the vault as soon as the read ends. A dropped connection, a 429 or a 5xx answer is retried twice (after about 1 and 3 seconds). A record that still cannot be delivered is logged as a warning on the runner, and remains in the local audit file. Delivery never blocks or fails a scan.
Using vault credentials in scans
How DAST, API security, source code and container scans, source-code integrations, and SSH and WinRM host scans use a vault entry or a PAM reference, and which execution setting each needs.
Security model
Trust boundaries, what each component can see, how PAM credentials stay inside your network and out of logs and uploads, and the controls you have over what a runner may read.