SaaS providers decide how long your audit logs exist and which events you can see at all, and the answer usually depends on the licence you bought rather than on what an investigation needs. Native retention ranges from seven days to a year, some of the most useful events sit behind premium tiers, and data that expires before you export it cannot be recovered. This guide explains how SaaS audit log retention and licensing work, compares what major platforms document today, shows how to close gaps with export and streaming, and sets out the decisions a security team should make before the next incident. It is for SOC leads, detection engineers and SaaS owners, and it goes deeper on SaaS than the Log Management and Retention Guide.
Key takeaways
- Native retention is short and fixed. Google Workspace keeps audit log events for six months, GitHub keeps organisation audit events for 180 days and Git events for seven days, and Entra ID Free keeps sign-in and audit logs for seven days.
- Licences gate events, not just days. In Microsoft 365, Audit (Premium) adds longer retention, custom retention policies and extra properties, and in Slack the Audit Logs API is limited to the Enterprise plan.
- Upgrades are not retroactive. Microsoft states that retention is set when a record is ingested and that expired Entra data cannot be recovered unless it was archived. Buy or export before you need it.
- Export is the real retention control. Streaming or polling into a SIEM or storage account you own is the only way to choose your own period and to search across SaaS platforms together.
- The collector is a privileged NHI. Every export uses an API token, OAuth app or service principal with read access to the whole tenant's activity, so it needs the same controls as any high-value credential.
Why retention and licensing decide investigation outcomes
Breach investigations routinely start weeks or months after initial access. If logs have aged out, the questions that matter (when did the attacker arrive, what did they read, which tokens did they mint) have no answer, and notification decisions have to assume the worst. Licensing makes it worse in two ways.
- Depth: lower tiers may omit whole event types. The Microsoft Storm-0558 key breach was spotted by US State Department analysts using an alert on the MailItemsAccessed log, available to them through premium audit logging. Organisations without it could not see the same activity, and Microsoft later expanded logging for all customers.
- Duration: even when an event is logged, a short window means slow-burn activity such as token abuse by an integration may only be visible in its last few weeks.
- Access: some platforms show recent events in the console but expose older events or certain categories only through an API or export feature that has its own tier requirements.
What major platforms document today
The table summarises vendor documentation checked on 27 October 2026. Vendors change tiers and periods, so record the date you checked and re-check yearly.
| Platform and log | Native retention (as documented) | Licence or tier notes | Export route |
|---|---|---|---|
| Microsoft Purview Audit (Standard) | 180 days for all records (records generated before 17 October 2023 were kept 90 days) | No custom retention policies | Office 365 Management Activity API, baseline 2,000 requests per minute |
| Microsoft Purview Audit (Premium) | 1 year for Exchange, SharePoint, OneDrive and Entra ID records; 180 days for others | Custom policies up to 1 year, or up to 10 years with the 10-Year Audit Log Retention add-on; extra properties such as MailItemsAccessed SensitivityLabel; intelligent insights | Same API with about twice the bandwidth |
| Microsoft 365 non-user records | Fixed one year in both tiers for service principals, system events and applications | Not configurable | Same API |
| Microsoft Entra ID sign-in and audit logs | Free: 7 days; P1 and P2: 30 days | Purview Audit (Premium) can retain Entra audit logs longer | Diagnostic settings to Log Analytics, storage or Event Hubs |
| Google Workspace audit and investigation logs | Six months for Admin, login, Drive, Gmail, OAuth Token, SAML, Groups and Rules | Administrators cannot delete log data or change the period; BigQuery export limited to listed Enterprise, Frontline and Education editions | Reports API; daily BigQuery export |
| GitHub organisation audit log | 180 days for web events; Git events seven days | Enterprise audit log streaming is an enterprise feature | Streaming to Amazon S3, Azure Blob Storage, Azure Event Hubs, Datadog, Google Cloud Storage or Splunk; REST API |
| Salesforce Setup Audit Trail | 180 days, after which setup entity records are deleted | Event Monitoring log files are a separate capability; check your contract | Download or API |
| Slack audit logs | Not stated on the API page | Audit Logs API available only to Enterprise plan organisations, through an app installed by the org Owner with the auditlogs:read scope | Audit Logs API |
For platform specific detection guidance see the Microsoft 365 Unified Audit Log Guide for Detection, the Entra ID Sign-In and Audit Log Detection Guide and the Google Workspace Audit Log Detection Guide.
Deciding how long to keep SaaS logs
Set the target from investigation and regulatory need, then compare it with what each platform keeps natively. The gap is what export must cover.
| Driver | Typical requirement | Source to check |
|---|---|---|
| Baseline control framework | CIS Controls v8.1 Safeguard 8.10 asks for at least 90 days of audit log retention | CIS Controls v8.1 |
| Federal guidance on event logging | NIST SP 800-53 AU-11 requires a defined retention period consistent with records retention policy | NIST SP 800-53 Rev. 5 |
| Payment card environments | PCI DSS v4.0.1 Requirement 10.5.1: retain audit log history for at least 12 months, with the most recent three months immediately available | PCI Security Standards Council |
| Incident investigation reality | 12 months searchable is a common internal target because many intrusions are found late | Internal risk decision |
| Contract and sector rules | Varies; financial and public sector contracts often specify longer periods | Legal and compliance |
Hot, warm and cold tiers
- Hot (30 to 90 days): full events in the SIEM for alerting and fast search.
- Warm (up to 12 months): lower-cost searchable storage such as a data lake or basic log tier, for investigations and threat hunting.
- Cold (beyond 12 months): immutable object storage with legal hold for regulatory and litigation needs.
Buy the tier or build the export
For each platform there are two ways to close a gap: pay for a higher tier, or export events to storage you control. They solve different problems.
| Question | Upgrade the licence | Export and retain yourself |
|---|---|---|
| Are events missing entirely on the current tier? | Only an upgrade adds them | Export cannot capture what is never logged |
| Are events logged but kept too briefly? | Works, but limited to vendor maximums | Usually cheaper and fully under your control |
| Do you need cross-platform correlation? | No help | Required |
| Do you need native investigation tools on old data? | Yes, in-product search over the longer period | Only through your own tooling |
| Small organisation with one main suite | Often the simplest path, for example a suite tier that includes premium audit | Consider a low-cost storage export as a backstop |
| Large organisation with many SaaS platforms | Upgrade where events are missing | Central pipeline is the default for everything else |
A practical rule: upgrade where the higher tier adds event types you need for detection, and export everywhere else. Record the decision per platform in the Log Source Coverage Matrix Template.
Where identity fits: the collectors
Every export path is a non-human identity: an Entra service principal reading the Management Activity API, a Google service account with domain-wide delegation for the Reports API, a Slack app with auditlogs:read, or a GitHub stream with credentials to a cloud bucket. Each one can read the activity of the whole tenant, and some can write to your storage.
- Scope narrowly: grant only read access to audit data, never a broader admin integration.
- Prefer federation: use workload identity federation or managed identities instead of long-lived secrets where the platform supports it.
- Own and inventory: each collector needs a named owner, a documented purpose and a rotation date. See the SaaS and OAuth App Governance Guide.
- Alert on change: a disabled stream or a deleted collector is a defence impairment action and should alert like one.
Questions to settle at contract and renewal
Retention and log access are cheapest to fix when a SaaS contract is signed or renewed. Put these questions to every tier-one vendor and keep the answers with the inventory.
- Event scope: which administrative, authentication, data access and API events are logged on the tier we are buying, and which require a higher tier or an add-on?
- Integration activity: are actions by OAuth apps, API tokens and service accounts logged with the client or token identifier, or only attributed to the user who authorised them?
- Retention: how long is each category kept, can we change it, and is the period applied at ingestion or at query time?
- Export: is there push streaming, a pull API, or only a console download? What are the rate limits, and what happens to events during an outage of our receiver?
- Format and fields: are events documented with a schema, stable identifiers and source IP addresses, and are timestamps in UTC?
- Tamper evidence: can tenant administrators delete or edit audit events, and is that action itself logged?
- Incident support: will the vendor provide logs beyond native retention during an investigation, and within what time?
If the answer to the export question is "console download only", treat the platform as a gap in the coverage matrix and plan compensating controls, such as logging at the identity provider and at a cloud access security broker, until the vendor improves. For a wider view of SaaS controls see the SaaS Security Guide.
Implementing a SaaS log retention programme
- Inventory platforms: list every SaaS platform that holds sensitive data or admin power. For each, record the current licence, native retention, the event categories available and the export options, with a link to the vendor page and the date checked.
- Set the target: agree retention targets per data class with legal and compliance, normally 90 days hot and 12 months searchable as a minimum.
- Close missing-event gaps first: identify platforms where the current tier lacks events your detections need (mail access, API access, file downloads) and budget the upgrade.
- Build the pipeline: connect the top platforms to the SIEM or data lake through streaming where available and polling otherwise. Use overlapping poll windows and de-duplicate on event ID.
- Pilot and validate: for two weeks, compare daily event counts in the SIEM with counts in each platform's console. Investigate any difference over a few percent.
- Protect the store: immutable or write-once storage for the cold tier, separate administrative accounts for the log platform, and access logging on the log store itself.
- Rollback and resilience: keep native retention as the fallback, document how to re-pull missed windows within the platform's retention, and know each platform's buffer (for example GitHub keeps a seven-day buffer for a paused stream).
- Review annually: vendors change tiers and periods. Re-check each platform's documentation and update the inventory.
Detection and measurement
- Coverage: percentage of in-scope SaaS platforms exporting audit events, target 100 percent of tier-one platforms.
- Retention gap: number of platforms where searchable history is shorter than the agreed target.
- Freshness: maximum ingestion lag per platform, with an alert when a source is silent beyond its documented lag.
- Completeness: daily count reconciliation between source and SIEM, as a percentage.
- Tier fitness: number of detection use cases blocked by missing events on the current licence.
- Collector hygiene: collectors with a named owner, least-privilege scope and current credential rotation.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev. 5 | AU-11 Audit Record Retention; AU-4 Audit Log Storage Capacity; AU-9 Protection of Audit Information | Retention period, capacity and protection of exported logs |
| NIST SP 800-53 Rev. 5 | AU-2 Event Logging; AU-6 Audit Record Review, Analysis, and Reporting | Choosing events and using them |
| NIST CSF 2.0 | PR.PS-04, DE.CM-09 | Log records are generated and made available; computing environments are monitored |
| CIS Controls v8.1 | Control 8 Audit Log Management, Safeguards 8.9 and 8.10 | Centralise and retain audit logs |
| ISO/IEC 27001:2022 | Annex A 8.15 Logging; 5.23 Information security for use of cloud services | Logging and cloud service agreements |
| MITRE ATT&CK v20 | T1685 Disable or Modify Tools | Attackers turning off auditing or streams |
Common failure patterns
- Assuming the console shows everything: some categories are API or export only, and some only appear on higher tiers.
- Buying the tier after the incident: Microsoft documents that retention is set at ingestion and that expired Entra data cannot be recovered. A licence bought during an investigation does not bring back last quarter.
- Retaining logs but not the right events: the Storm-0558 case showed that the difference between detecting and missing a mailbox intrusion was one event type available on one tier.
- Ignoring integration activity: logs of service principal and OAuth app activity are the ones that show token abuse. In the Salesloft Drift breach stolen integration tokens were used to query Salesforce data, so API and integration logs, and enough history to scope them, decide how quickly victims can answer what was taken.
- Silent pipeline failures: expired collector credentials or API throttling stop ingestion without anyone noticing until an investigation finds a gap.
- Short-lived Git and API events: GitHub Git events are kept for seven days natively. Without streaming, clone and push history is gone within a week.
Practitioner checklist
- Inventory every tier-one SaaS platform with licence, native retention, available events and export route, dated.
- Agree retention targets with legal: at least 90 days hot and 12 months searchable.
- Upgrade tiers where events needed for detection are missing; export everywhere else.
- Stream or poll audit events into storage you control before they expire natively.
- Run each collector as a least-privilege, owned and inventoried non-human identity.
- Reconcile daily event counts between source and SIEM.
- Alert on silent sources and on disabled streams or audit settings.
- Store the cold tier immutably with separate admin access.
- Re-check vendor retention and licensing documentation every year.
Standards and references
- Microsoft Learn: Learn about auditing solutions in Microsoft Purview
- Microsoft Learn: Microsoft Entra data retention
- Google Workspace Admin Help: Data retention and lag times
- GitHub Docs: Reviewing the audit log for your organization
- GitHub Docs: Streaming the audit log for your enterprise
- Slack Developer Docs: Audit Logs API
- Salesforce Help: Monitor Setup Changes with Setup Audit Trail
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Log Management and Retention Guide · CIS Control 8 Audit Log Management Guide · Microsoft 365 Unified Audit Log Guide for Detection · SIEM Guide