Google Cloud always writes Admin Activity logs, but the logs that show who read data, who minted a token and who decrypted a key are Data Access logs, and they are off by default for every service except BigQuery. Out of the box, project logs also stay in the project for 30 days (or 400 days for the protected _Required bucket), where the project owner can reroute or exclude much of them. This guide covers the four Cloud Audit Logs types, how to enable Data Access logging at organisation level without a runaway bill, how to centralise with aggregated and intercepting sinks as the Google Cloud enterprise foundations blueprint does, how to lock retention, and what to detect. It is for cloud security engineers, detection engineers and auditors. For logging principles across providers, read the Cloud Logging and Audit Trail Guide.
This is an independent guide from NHI Mgmt Group, based on Google Cloud's public documentation as of November 2026. It is not sponsored or reviewed by Google Cloud. Product features change, so confirm settings against Google Cloud's current documentation.
Key takeaways
- Four types, two always on. Admin Activity and System Event logs cannot be disabled or excluded; Policy Denied logs cannot be disabled but can be excluded; Data Access logs are off except for BigQuery.
- Enable Data Access by design, not everywhere. Configure
auditConfigsat the organisation for high-value services (IAM, Cloud KMS, Secret Manager, Cloud Storage in sensitive folders), because child projects cannot disable inherited settings. - Centralise with aggregated sinks at the organisation. Use
--include-children, or--intercept-childrenwhen project owners must not be able to drop logs before they reach the central project. - Retention is not long enough by default.
_Requiredkeeps 400 days and_Default30 days; route to a locked log bucket or Cloud Storage with Bucket Lock for compliance retention. - The sink is a target. Alert on sink deletion or filter changes, exclusion additions and
auditConfigsedits, which map to ATT&CK T1685.002 Disable or Modify Cloud Log.
How identity and access work in Google Cloud Audit Logs
Every audit entry records the authenticated principal and the method called. The four log types differ in what triggers them and whether you can turn them off.
| Log type | Log name suffix | What it records | Configurable | Default bucket |
|---|---|---|---|---|
| Admin Activity | cloudaudit.googleapis.com%2Factivity | Calls that change configuration or metadata (ADMIN_WRITE), such as SetIamPolicy | No: always written, cannot be excluded | _Required |
| System Event | cloudaudit.googleapis.com%2Fsystem_event | Changes made by Google systems or service agents | No | _Required |
| Data Access | cloudaudit.googleapis.com%2Fdata_access | ADMIN_READ (metadata reads), DATA_READ and DATA_WRITE (user data) | Yes; off by default except BigQuery, whose Data Access logs cannot be disabled | _Default |
| Policy Denied | cloudaudit.googleapis.com%2Fpolicy | Access denied by a security policy, such as VPC Service Controls | Cannot be disabled; can be excluded | _Default |
Fields that matter in an investigation
protoPayload.authenticationInfo.principalEmail: the user or service account that called the API. For service accounts, also check delegation information to see whether a human impersonated it.protoPayload.methodNameandserviceName: the operation, such asSetIamPolicyoncloudresourcemanager.googleapis.com.protoPayload.requestMetadata: caller IP and user agent, the main signal for credential use from an unexpected location.protoPayload.authorizationInfo: which permission was checked and whether it was granted.resourceandlogName: the monitored resource type and the project, folder or organisation the entry belongs to. A filter oflogName:"cloudaudit.googleapis.com"selects all audit logs.
Admin roles and privileged access
Access to logs is itself privileged: Data Access logs can contain resource names and request details, and whoever controls sinks controls what your SOC sees.
| Role | Role ID | What it allows | Risk if over-granted |
|---|---|---|---|
| Logs Viewer | roles/logging.viewer | Read Admin Activity, System Event and Policy Denied logs; not Data Access in _Default | Low; exposes configuration history |
| Private Logs Viewer | roles/logging.privateLogViewer | Adds Data Access logs and all logs in _Required and _Default | Reveals who accessed which data; restrict to SOC and auditors |
| Logs Configuration Writer | roles/logging.configWriter | Create and change sinks, buckets, exclusions | Can silently stop routing (defence impairment) |
| Logging Admin | roles/logging.admin | All Logging permissions | Full control of the log pipeline |
| Owner on the logging project | roles/owner | Grant writer identities, delete buckets (unless locked) | Single point of failure for evidence |
Anyone with setIamPolicy on the organisation | Organization Administrator and similar | Edit auditConfigs, which live in the IAM policy | Can turn Data Access logging off fleet-wide |
- Editor is not enough, by design: Google notes the Editor role does not include permission to read Data Access logs, so investigators need Private Logs Viewer explicitly.
- Log views for least privilege: reading logs from a bucket requires access to a log view, and views can expose a subset of a bucket, for example one application team's projects. Grant on views rather than the whole central bucket.
- Separate the logging project: the blueprint keeps all log storage in a dedicated project (
prj-c-logging). Keep workload teams out of it and restrict Owner to a break-glass group. See the Google Cloud Enterprise Foundations Blueprint Guide.
Integrations, API tokens and service accounts
Enabling Data Access logs
Data Access logging is configured in the auditConfigs section of an IAM policy, at organisation, folder, project or billing account level. Each entry names a service or allServices and the log types to enable. Configurations are unioned down the hierarchy: a project can add log types but cannot remove ones inherited from its folder or organisation.
| Service | Log types to enable | Why | Volume |
|---|---|---|---|
iam.googleapis.com, iamcredentials.googleapis.com | ADMIN_READ, DATA_READ | Service account token generation and impersonation, key listing | Low to medium |
cloudkms.googleapis.com | DATA_READ, ADMIN_READ | Encrypt, Decrypt and Sign are DATA_READ; shows who used a key | High for busy CMEK services |
secretmanager.googleapis.com | DATA_READ | Secret version access | Medium |
storage.googleapis.com | DATA_READ, DATA_WRITE in sensitive folders | Object reads (T1530 Data from Cloud Storage) | Very high; scope by folder |
cloudresourcemanager.googleapis.com | ADMIN_READ | IAM policy reads during reconnaissance | Medium |
| BigQuery | On by default | Query and table reads | Already collected |
- Edit safely: use read-modify-write with
getIamPolicyandsetIamPolicy, keep theetag, and never drop thebindingsblock, or the write removes all access. - Exempted principals carefully: you can exempt principals per service and log type to cut noise, for example a backup service account reading Cloud Storage. Exemptions are a blind spot: document each one and review them. Adding an exemption to a service also enables that log type.
- Default configuration: an organisation-level default applies automatically to services that start producing Data Access logs later.
Sink writer identities
Each sink writes through a Google-managed service account, its writer identity, shown in gcloud logging sinks describe. Since 22 May 2023 Logging uses one service account for all sinks in the same resource. Grant it only the destination role it needs: roles/logging.bucketWriter for a log bucket, roles/storage.objectCreator for Cloud Storage, roles/bigquery.dataEditor for BigQuery, roles/pubsub.publisher for Pub/Sub. If the destination is in a VPC Service Controls perimeter, add an ingress rule for that service account. See the VPC Service Controls Guide.
SIEM connectors that pull from Pub/Sub or BigQuery use their own service accounts. Use workload identity federation instead of exported keys where the SIEM supports it, and grant subscriber access on one subscription only. See the Google Cloud Service Account Keys Security Guide.
Logging and detection
Routing architecture
| Component | Behaviour | Design choice |
|---|---|---|
_Required sink and bucket | Admin Activity, System Event and Access Transparency logs; cannot be modified or deleted; 400-day retention, not changeable | Your floor, not your archive |
_Default sink and bucket | Everything else; can be modified or disabled; 30 days (project buckets configurable 1 to 3650 days) | Keep for local troubleshooting; set organisation default resource settings for new projects |
| Aggregated non-intercepting sink | Copies logs from all children (includeChildren) to a central destination; children still route locally | Default for central security copy |
| Aggregated intercepting sink | interceptChildren: logs are routed centrally and not processed by child sinks | When local copies are unwanted or project owners must not reroute |
| Log bucket with Log Analytics | SQL queries, linked BigQuery dataset; region fixed at creation; upgrade irreversible | Active investigation (blueprint sk-c-logging-la) |
| Cloud Storage bucket | Long-term archive; add Bucket Lock where retention is mandatory | Compliance retention (blueprint sk-c-logging-bkt) |
| Pub/Sub topic | Stream to an external SIEM | SIEM feed (blueprint sk-c-logging-pub) |
Limits to plan for: 200 sinks per folder or organisation, 50 exclusion filters per sink, filters up to 20,000 characters, and global service endpoints only for destinations. Sinks never route retroactively, so a gap caused by a broken sink stays a gap. Folder and organisation buckets cannot have retention extended: route their logs to a project bucket.
Detections to build
| Behaviour | Signal (log type, method) | ATT&CK |
|---|---|---|
| Sink deleted, filter changed or exclusion added | Admin Activity, Logging config methods (DeleteSink, UpdateSink, exclusion create) | T1685.002 Disable or Modify Cloud Log |
| Data Access logging reduced | Admin Activity, SetIamPolicy with changed auditConfigs | T1685.002 |
| Owner, Editor or Organization Admin granted | Admin Activity, SetIamPolicy binding deltas (the blueprint uses a Cloud Asset Inventory feed for this) | T1098.003 Additional Cloud Roles |
| Service account key created | Admin Activity, IAM CreateServiceAccountKey | T1098.001 Additional Cloud Credentials |
| Token minted by impersonation | Data Access on iamcredentials.googleapis.com (GenerateAccessToken) | T1078.004 Cloud Accounts |
| Bulk object reads | Data Access DATA_READ on Cloud Storage | T1530 Data from Cloud Storage |
| VPC Service Controls denials | Policy Denied logs | Exfiltration attempts against a perimeter |
The blueprint's bq-log-alerting sample runs BigQuery views over the central logs every 15 minutes and raises custom Security Command Center findings for events such as super admin logins, logging setting changes and route changes. See the Google Security Operations (Chronicle) Guide for SIEM-side analytics.
Metrics that show it is working
- Coverage: percentage of projects whose audit logs reach the central sink, checked by a daily query of distinct project IDs against Cloud Asset Inventory.
- Data Access coverage: number of high-value services with Data Access enabled at organisation or folder level versus the target list.
- Pipeline integrity: unapproved sink, exclusion or
auditConfigschanges per quarter (target zero). - Latency: median time from event to SIEM availability.
- Retention compliance: percentage of required log classes held in locked storage for the mandated period.
Hardening baseline
- Central logging project: one project for all log storage, outside workload folders, with Owner limited to break-glass and a project lien against deletion.
- Organisation-level aggregated sinks: at least one non-intercepting sink for all audit logs (
logName:"cloudaudit.googleapis.com") to a Log Analytics bucket and one to Cloud Storage or Pub/Sub for archive and SIEM. - Data Access by service list: enable the services in the table above at organisation or folder level, scoped to sensitive folders for high-volume services.
- Locked retention: lock the central log bucket (irreversible; gcloud or API only) or apply Bucket Lock on the archive bucket for the mandated period.
- Default resource settings: set organisation defaults for the
_Defaultsink so new projects behave consistently. - Network logs: the blueprint also enables VPC Flow Logs (50 percent sampling), Firewall Rules Logging and Cloud DNS logs; new subnets, rules and zones need enabling as they appear.
- Encryption and perimeter: use CMEK for log buckets where policy requires it, and place the logging project in a VPC Service Controls perimeter. See the Cloud KMS and Cloud HSM Guide.
Phased rollout
- Measure: list existing sinks at every level, current
auditConfigs, and per-project log volume. - Central sinks first: create the logging project and organisation-level non-intercepting sinks; grant writer identities; confirm entries from a sample of projects arrive.
- Data Access in a pilot folder: enable the target services, measure volume for two weeks, add documented exemptions only where volume is unjustified.
- Extend: move the configuration to organisation level for low-volume services and to sensitive folders for high-volume ones.
- Lock: once retention and region are agreed, lock the bucket. Region and lock cannot be changed afterwards, so test on an unlocked bucket first.
- Detect: deploy the detections above and test each by making a harmless change in a test project.
- Roll back: remove a service from
auditConfigsat the level you set it, or add an exclusion filter (create it disabled, then enable) on a sink. Locked buckets cannot be rolled back.
Decision guidance
- Intercepting vs non-intercepting: intercepting gives stronger central control and avoids double storage, but removes logs teams use locally. Most organisations start non-intercepting and intercept only for regulated folders.
- Log bucket vs Cloud Storage vs Pub/Sub: log buckets for search, Cloud Storage for cheap locked archive, Pub/Sub for an external SIEM. Large estates typically use all three, as the blueprint does.
- Small organisations: one organisation sink to a Log Analytics bucket with 400-day retention and Data Access for IAM, KMS and Secret Manager is a reasonable minimum.
Framework mapping
| Control area | NIST SP 800-53 Rev 5 | NIST CSF 2.0 | CIS Controls v8.1 | ISO/IEC 27001:2022 Annex A |
|---|---|---|---|---|
| Audit log generation and Data Access scope | AU-2, AU-12 | PR.PS-04 | 8 Audit Log Management | A.8.15 Logging |
| Central collection and review | AU-6, SI-4 | DE.CM-09, DE.AE-02 | 8 Audit Log Management | A.8.16 Monitoring activities |
| Protection of logs, locked buckets, sink integrity | AU-9 | PR.PS-04 | 8 Audit Log Management | A.8.15 Logging |
| Retention | AU-11, AU-4 | PR.PS-04 | 8 Audit Log Management | A.5.28 Collection of evidence |
| Access to logs | AC-6 | PR.AA-05 | 6 Access Control Management | A.8.2 Privileged access rights |
Common misconfigurations
- Assuming Data Access is on: teams investigate a suspected secret read and find only Admin Activity logs, because Secret Manager and IAM credentials Data Access logging was never enabled.
- Enabling allServices DATA_READ everywhere: the bill spikes, someone turns it off at the organisation, and coverage drops to nothing. Enable per service and folder.
- Project-level sinks only: each team configures its own export, new projects are missed, and a project owner can delete the sink. Use organisation-level aggregated sinks.
- Unreviewed exemptions and exclusions: an exempted service account or a broad exclusion filter becomes a permanent blind spot attackers can exploit if they compromise that principal.
- Relying on _Required for retention: 400 days covers Admin Activity only; Data Access and Policy Denied logs in
_Defaultare gone after 30 days unless routed. - Locking too early: a bucket locked in the wrong region or with the wrong retention cannot be changed or deleted until entries expire.
Practitioner checklist
- Create a dedicated logging project with restricted Owner and a project lien.
- Deploy organisation-level aggregated sinks for all audit logs to a Log Analytics bucket and an archive or SIEM destination.
- Grant each sink writer identity only the destination role it needs, and add VPC Service Controls ingress rules where required.
- Enable Data Access logs for IAM, IAM Credentials, Cloud KMS and Secret Manager at organisation level, and for Cloud Storage in sensitive folders.
- Document and review every exempted principal and exclusion filter quarterly.
- Lock retention for the mandated period after testing region and retention on an unlocked bucket.
- Restrict Private Logs Viewer, Logs Configuration Writer and Logging Admin, and use log views for team access.
- Alert on sink, exclusion and
auditConfigschanges, privileged role grants and service account key creation. - Check daily that every project appears in the central logs.
Standards and references
- Google Cloud: Cloud Audit Logs overview
- Google Cloud: Enable Data Access audit logs
- Google Cloud: Aggregate and store your organization's logs
- Google Cloud: Routing and storage overview
- Google Cloud: Configure log buckets
- Google Cloud: Cloud Logging quotas and limits
- Google Cloud: Enterprise foundations blueprint, detective controls
- Google Cloud: Understand audit logs
Related NHI Mgmt Group resources: Cloud Logging and Audit Trail Guide · Google Cloud Security Guide · Google Cloud Enterprise Foundations Blueprint Guide · Google Cloud IAM and Service Accounts Guide · SIEM Guide