2026.11
| Version | |
|---|---|
| 2026.11.0 | November 17, 2026 |
System Requirements Update
Ensure your infrastructure meets the latest requirements for optimal performance and security:
| Component | Minimum Version | Recommended Version |
|---|---|---|
| KOTS | 1.117.3 | Latest |
| Kubernetes | 1.30 | 1.36 |
| PostgreSQL | 15 | 17 |
| Redis | 6 | 7 |
| ggscout | 0.32.0 | Latest |
The chart ships ggscout 0.35.0. Validity and permissions checks need ggscout 0.34.1 or later, so upgrade any ggscout you run outside the cluster to get them.
Helm & Upgrade Considerations
To ensure compatibility, please review Helm values updates from the previous version. Air gap deployment? Find all the images and tag names in the air gap install page.
⏳ Background migrations: an upgrade to 2026.11 is refused with
Not all required migrations have been appliedwhile background data migrations started in 2026.8 and 2026.9 are still running. Instances that ran 2026.10 long enough have usually finished them. If the upgrade is refused, wait for them to finish on your current version, then retry.
Helm global values
On Helm installations, more global.* values now reach the bundled components:
- ClickHouse and Keeper read
global.priorityClassName,global.tolerations,global.podAnnotationsandglobal.labels.global.nodeSelectorreaches them only when they have no node placement of their own. - The log collector's MinIO (
loki-minio) readsglobal.priorityClassName,global.nodeSelector,global.tolerations,global.podAnnotationsandglobal.labels.
If you set these values, ClickHouse, Keeper and MinIO pods restart once during the upgrade and can be scheduled on other nodes. A volume already bound to a node that a new selector excludes keeps the pod Pending.
The Replicated SDK (replicated) and Loki (loki) still ignore most global.* values. For example, with global.tolerations set, their pods can still stay Pending on a tainted node. Set their own keys instead, such as replicated.tolerations and loki.singleBinary.tolerations. The description of each global.* value in the Helm values reference names the key to use.
Feature highlights
- Risk score: turn on the machine learning risk score in the Admin Area preferences, and every secret incident gets a score from 0 to 100, with the Critical saved view listing the open incidents scored 85 or above, so you know which ones to remediate first. The risk score is off by default. Once it is on, new incidents are scored as they come in, and the risk score backfill scores your past incidents. The scanner workers compute the score unless you run the dedicated risk score worker. Learn more.
- AWS S3 on Self-Hosted: scan your S3 buckets for secrets, with a read-only IAM role that GitGuardian assumes through OIDC: no long-lived AWS credentials. The integration and historical scans are on by default. AWS must reach your instance over the internet to validate its tokens, so air-gapped instances can't use it. Learn more.
- Harbor container registry (beta): scan every layer of your tagged Harbor images with a read-only robot account, and pick the repositories to monitor or let new ones be added automatically. Learn more.
- Validity and permissions of your vaulted secrets: ggscout checks every secret it collects with its provider, from inside your network, and the NHI inventory now shows a Validity column, with permissions, owner and expiry in the identity detail. Learn more.
Secrets Detection Engine
- v2.174: 10 new detectors (Lago API Key, Composio API Key and Organization API Key, Codex API Key, CARTO Basemaps API Key, Google Chat Webhook URL, SFTP Credentials, HubSpot Personal Access Key, AmneziaWG Private Key, Kubernetes Kubeconfig Bearer Token), 31 detector updates (SMTP credentials checker enabled, custom host and proxy support for the Azure Entra, Azure Active Directory, GitLab Deploy Token and LiteLLM checkers, fewer false positives for Generic High Entropy Secret and Generic Database Assignment, the
_with_hostvariants merged into their base detectors), 2 new analyzers (GoCardless, Adobe), 1 analyzer update (GitLab Token), 1 deprecated detector (LINE Notify Token). - v2.175: 1 new detector (Portainer API Key), 2 new checkers (Weights & Biases API Key, ODBC Connection String), 8 checkers available with a custom host (Artifactory Token and Reference Token, AWS Cognito OAuth 2.0 Credentials, Bitbucket Data Center HTTP Access Token, Jira Data Center Personal Access Token, Microsoft Azure Storage Account Key, Okta Keys and Token), fewer false positives for Generic Password and Generic High Entropy Secret, 1 new analyzer (Weights & Biases), 1 analyzer update (Google Cloud Keys).
Enhancements
- The progress counters of running bulk scans refresh more often. Learn more.
- Findings pushed to Wiz carry the GitGuardian incident, occurrence and source IDs. Learn more.
- Machine Scan on Windows covers fixed drives only, scans skip more folders of downloaded or generated files, and validity checks take custom hosts only from your workspace. Learn more.
- Validity checks go through GitGuardian Bridge for detectors that read the host from the secret, custom webhooks accept several custom headers, the public API can unlink a ticket from an incident and return the content of an occurrence in clear text, and
ggshield auth loginasks for the Endpoint Protection scopes separately. Learn more. - Password logins are locked out for six minutes after ten failed attempts for the same email and IP address, and an audit log entry records each lockout. The public API filters GitHub check runs by commit SHA, pull request number or source, an integration's perimeter can select every source that matches the filters, and federated JFrog Docker repositories are scanned. Learn more.
- Self-Hosted:
- Jira Cloud and Confluence Cloud historical scans now use the DB-less scan framework by default, which turns on the DB-less scan flag of each integration in the Admin Area, even where you had turned it off. Both scans keep running on your
worker-scanners-odspods if you run them, or onworker-scannersotherwise. To run them on dedicated pods, setceleryWorkers.scanners-db-less-jira-cloud.replicasandceleryWorkers.scanners-db-less-confluence-cloud.replicas, or the matching worker options in the KOTS Admin Console. - ClickHouse can now use its own encryption keys, separate from the PostgreSQL ones, set with
miscEncryption.existingSecretKeys.clickhouseEncryptionKeys. A dedicated key keeps ClickHouse data safe if the PostgreSQL key leaks. When unset, ClickHouse keeps using the PostgreSQL keys, so there is nothing to do on upgrade. The Admin Area key rotation page now rotates each database's keys separately. See Database security for the risks of a single key and how to set a dedicated one. - Google Artifact Registry integrations move to the new version of the integration in the background, which adds keyless connection with Workload Identity, flags unhealthy installations and lets you update a service account key without reinstalling. The migration must finish before you can upgrade to 2027.4.
- Set the password login lockout in the Admin Area preferences:
max_login_failed_attempts(0turns it off) andlogin_lockout_duration. A password reset lifts a user's lockout. - A backfill of past incidents that makes no progress for 24 hours is marked as stopped, so that you can start a new one. Change the delay with the
backpopulation_expiry_hourspreference. - The containers that had no resource settings, such as the machine learning engine's
installinit container, now ship with requests and limits that you can change, so a namespace with a strictResourceQuotaadmits every GitGuardian pod. See the Helm values changelog. - The usage data your instance sends now covers remediation and analytics KPIs, the incident grouping setting, the ClickHouse cache hit ratio and the time collection takes.
- Helm
global.*values reach ClickHouse, Keeper and the log collector's MinIO. See Helm & Upgrade Considerations.
- Jira Cloud and Confluence Cloud historical scans now use the DB-less scan framework by default, which turns on the DB-less scan flag of each integration in the Admin Area, even where you had turned it off. Both scans keep running on your
Fixes
- The historical scan report email links to the scan it reports on. Learn more.
- Confluence Cloud incidents found in attachments show the user who uploaded them, a rejected Confluence Data Center personal access token is reported as invalid instead of raising an internal error, and a token of a blocked GitLab account is reported as such instead of as GitLab being unreachable. Learn more.
- A member provisioned by SCIM no longer hits an error on their first SAML sign-in, Google Artifact Registry integrations with a monitored package are no longer reported as unhealthy, and an incident assigned with a service account token sends the assignment email. Learn more.
- Tagging a filtered selection of public incidents with select all no longer tags every public incident, and SharePoint Online and OneDrive no longer require the
Organization.Read.Allpermission on your Entra application. Learn more. - Self-Hosted:
- Your instance no longer connects to the GitGuardian plugin repository every 10 minutes when no workspace uses the Endpoint Protection plugin catalog. See network requirements.
- Batched tasks, such as validity checks, no longer hold
worker-longpods until their batch is flushed, which delayed the other long-running tasks on large instances. They now run on aworker-batch-task-workerpod that runs by default, with 1 replica that autoscales up to 5. See Scaling. - Worker pods no longer try to reach the external LangSmith API when no tracing is configured.
- Long exports on large workspaces no longer time out after 5 minutes: the API pods that serve long requests now allow 15 minutes, and also serve the analytics pages, which could time out on large workspaces.
- The GitHub Enterprise integration page no longer freezes or crashes on perimeters with hundreds of thousands of repositories.
- Security: email and password authentication is now accepted only to sign in to the dashboard. Scripts that called other endpoints with an email and password must use an API token instead.