Astralis 2.95.0 release notes

• 2.95.0: September 29, 2026

Astralis 2.95.0

Data Mapping

Detect and classify global data risks with comprehensive visualization for data mapping and reporting for modern Enterprise.

• Reuse Existing Systems From The Action Center: When you add a vendor from the Action Center and a system with the same name already exists, you are offered that system instead of creating a duplicate. Choosing it links the system to the vendor and moves the vendor's pending resources onto it. Website and identity provider monitors resolve a vendor's system within the monitor's own organization, and Action Center cards group by the system a user assigned rather than only the one attribution found. Seeded sample systems carry their vendor, so website monitor results land on the sample system instead of creating a duplicate.
• Pre-Consent Detection For Google Consent Mode: Website monitor assets that fire Google Consent Mode as granted by default in consent-required regions are flagged as pre-consent, including when the same asset also fires after a denial. Notice-only notices no longer widen the Consent Mode storage an asset is judged against.
• Clearer TCF-Compliant Toggle: In the Action Center, the "TCF compliant" toggle is disabled with an explanatory tooltip on assets that are not on a TCF experience page, where marking compliance has no effect. Consent status details record whether each page and location served a TCF experience.
• Reprocess Website Monitor Results Without Rescanning: Website monitor results can be reprocessed from an already-completed scan without running a new one, optionally replacing the monitor's discovered assets outright rather than merging into them. Each scan's remote jobs are recorded against the monitor and task that submitted them as soon as they are submitted. Includes a database migration.
• Partial Monitor Configuration Updates: A new PATCH endpoint updates individual discovery monitor configuration attributes without resending the whole configuration.
• Dedicated Asset Report Permission: The Asset Report has its own asset_report:read scope, so it can be granted independently of the Data Map Report, which previously shared datamap:read. Every role and custom role holding datamap:read receives the new scope automatically, and the Admin UI "Asset report" navigation entry now requires it. See the upgrade note below for OAuth clients. Includes a database migration.
• Collision-Free Resource Identifiers For All Monitors: New monitors use the collision-free staged-resource URN encoding introduced in 2.94.0 from creation, and existing monitors are re-keyed by a post-upgrade backfill. See the rollback note below. Includes a database migration.
• Cross-Source Catalog Matching (In Development): Groundwork for matching a data catalog's entries to the datastore resources they describe. A data catalog monitor can be linked to the datastore monitors it should be matched against, and a matching job records when resources discovered by different monitors describe the same column or table. The UI is gated behind the alphaCatalogNormalization feature flag while it is in development. Includes database migrations.
• Bug Fixes:
  • Cloud infrastructure scans query each AWS region separately, so a resource is discovered once with stable tags and metadata instead of by both its home region and an aggregator region. Resources AWS reports as global are now scanned, and reconciliation is held back when a region could not be read so its resources are not mistaken for removed ones.
  • Identity provider apps from the same vendor are no longer merged into one system.
  • Website monitor discovered assets keep a stable order across refreshes, and the Action Center compliance column shows a consent status chip for every status.
  • Resource names in the discovery field path are decoded, and the resource tray shows the URN.
  • Long Action Center asset names truncate instead of pushing table columns out of view, and monitor statistics no longer stretch the datastore monitor page when a monitor has many data categories.
  • The system aggregate results endpoint no longer returns duplicate data uses for vendor-backed rows.
  • The Configure data use modal no longer spins forever on some declarations.
  • The Edit dataset YAML modal no longer overflows.

DSR Orchestration

Enterprise grade privacy request management and processing with a single orchestration layer for DSRs across vendors and systems

• Bulk Actions Across Every Matching Request: The Request Manager and the Flagged tab support a true select-all driven by the current filters, so bulk actions reach every matching request rather than only the loaded page. Very large selections run in lots with a progress modal and a Stop control instead of one call that can time out. The previous 10,000-row cap is replaced by the admin_ui.max_privacy_request_bulk_rows setting (default 25,000), and bulk responses report succeeded, failed, and truncation counts.
• Filter And Act On Suspicious Request Traffic: Privacy request search and every filter-driven bulk action accept risk signal filters: signal name, reason, score band, and enforcement mode at scoring. The Flagged tab can be filtered by reason, mode at scoring, score range, date, and source, and bulk Process or Delete can act on every request matching those filters. Includes a database migration.
• Skip Manual Tasks With No Context: Manual task configurations can skip creating a task when every configured context source is empty. The skip condition tracks the task's current context sources as they are added, changed, or removed, and saving a configuration now preserves context sources you did not include instead of clearing them.
• Access Package Download Link Destination: Access package download links can redirect through either the Privacy Center or the Fides API.
• Lower Memory Use For Polling Access Tasks: Polling access tasks store their aggregated results as a pointer manifest by default, cutting the memory peak during aggregation. Set FIDES__EXECUTION__POLLING_ACCESS_MANIFEST_ENABLED=false to restore the previous behavior.
• Alert On Unredacted Access Packages: An alertable warning and metric fire when an access package is generated with the default, non-redacting builder while the redacting access package builder was expected.
• Every Request Type In Policy And Rule Modals: DSR policy and rule modals offer every supported request type under one consistent label.
• Bug Fixes:
  • The DSR policies page and every policy picker load all policies instead of only the first 50, and a property action no longer reports a policy beyond the first page as "not found".
  • The privacy request CSV export honors the include_identities and include_custom_privacy_request_fields options instead of always including identity and custom field columns. See the upgrade note below.
  • Dangling dataset references are rejected when a dataset is linked to an integration, with an error naming each source field and its missing destination, instead of breaking every privacy request at runtime.
  • DSR workers build connectors from the same registry as the web server, so connectors are available consistently in every worker.
  • Worker shutdown is no longer retried or recorded as a privacy request failure.
  • SaaS responses where a single record is missing the filter field, or carries an empty list in it, no longer cause the whole collection to be discarded or the request to fail.
  • Hybrid manual task configurations follow template edits on upgrade, and every configuration type reads back correctly, not only access.
  • Manual fields can no longer be configured with a custom request type, and rule and configuration writes are validated.
  • Manual task digest test emails link internal users to the Admin UI instead of the external task portal.
  • The Jira tickets panel on a privacy request is visible to approvers, and Jira write actions are hidden from users who cannot perform them.
  • Privacy request activity timeline titles no longer collapse on crowded rows.
  • Temporal DSR worker logs carry the privacy request ID, so a failed request can be traced end to end.

Consent Management

Consent management and enforcement across web, mobile, and server-side surfaces

• Policy Documents (Alpha): The foundation for authoring and versioning privacy policies, privacy notices, and cookie policies, behind the alphaPolicyDocuments feature flag and new policy_document scopes. Three built-in document types ship with the release, and organizations can add their own. Documents keep an immutable version history, and saving detects when someone else has saved since you started editing. Documents are written in Markdown and rendered to sanitized HTML, and every outbound link is classified against the organization's own domains for review. Includes database migrations.
• Bug Fixes:
  • GPP no longer ignores saved consent for returning visitors who have no fides_string.
  • The iOS TCF consent experience is reachable by hardware keyboard, Full Keyboard Access, and VoiceOver. It opens in its own window and moves focus into it when shown.

Assessments

AI-assisted privacy assessments

• Assessment Findings And Evidence: Assessments have a detail view showing each requirement, whether it is satisfied, and the evidence behind that finding, alongside the risks the assessment surfaced. Statements, risks, and cited evidence are served from the assessment's own recorded findings. A satisfied finding must cite at least one fact, at most one artifact is current per assessment, and risks are superseded rather than deleted, so the history of an assessment's conclusions is preserved. Includes a database migration.
• Agentic Home Briefing: The Astralis Home briefing is written by a model from computed facts and is on by default. The briefing is stored and served for a configurable window, so the model runs once per window rather than on every read. This also fixes the Home briefing, which previously always failed to load.
• Assessments v2 Answers Become Facts (Alpha): Behind the alphaAssessmentsV2 flag, a submitted interview answer is written as a fact through a durable background workflow, with invalid submissions quarantined item by item. Writing a fact re-runs the known-gate, so a question that has just been answered stops being asked. Interview answers carry their own trust ceiling, data map fields that are adjacent evidence rather than an answer are recorded as partial coverage, and extracted facts are dated from the inventory's last edit. The gate decision ledger is scoped per organization. Includes a database migration.
• Framework Content Ratification (Alpha): A decomposition run can be reviewed and its drafts ratified as served framework content. Ratified content is locked and superseded rather than edited in place, records who ratified it and when, and requires an actor. Editing cited content clears its grounding verdict, and a replaced verdict records when it was asserted. Includes database migrations.
• Bug Fixes:
  • Privacy assessment re-evaluation judges context changes instead of never evaluating them.
  • Questionnaire and assessment intake turns no longer fail on unsupported tool calls.

PBAC

Access policies and purpose-based access control

• Policy Authoring In The v2.1 Policy Language: The policy chat agent and document-driven policy generation both author v2.1 policy language documents directly. Every draft is validated against the grammar and the organization's catalog, and validation errors are surfaced in the conversation or noted on the dropped draft with the fix it needs. The chat agent also flags drafts that would save but enforce nothing, such as a taxonomy key, notice, or dimension the organization does not have, a condition on an attribute nothing produces, or a priority outside the 0 to 100 convention. Document-driven generation now drafts with Claude Sonnet 4.6.
• Data Governance In Astralis: The Access policies pages, the Data purposes list and detail pages, the Data consumers list, create, and edit pages, and the Data governance Activity dashboard are available in the Astralis app under Data governance.
• Bug Fixes:
  • Violation details show the policy description and name the policy that raised each violation.

Integrations

• External Secrets Managers For Integration Credentials (In Development): Groundwork for integrations that read their credentials from an external secrets manager instead of storing them inline. The release adds an admin API and scopes for secrets manager configurations, cross-account access to AWS Secrets Manager through an assumed IAM role with a server-generated external ID, and a count of the integrations each manager backs. Only the owning organization can change or test a manager, bootstrap credentials are accepted but never returned, the test endpoint is throttled per manager, and deleting a manager that integrations still depend on is refused with those integrations named. Integrations do not resolve credentials through a secrets manager yet. Includes a database migration.
• AWS IAM Authentication For SaaS Connectors: SaaS connectors can authenticate with AWS Signature V4 signing, using static credentials or STS AssumeRole with automatic credential caching.
• Microsoft SQL Server Query Configuration And Schema Discovery: Microsoft SQL Server connections support query configuration and schema discovery.
• Bug Fixes:
  • A database password containing a space is no longer rejected at connection time. See the upgrade note below.
  • A deployment with no usable AWS identity is reported clearly by the secret provider health check instead of as an unexplained fetch failure, and a revoked KMS key clears the cached credential instead of serving it stale.
  • The integration detail page, add-integration wizard, and manage-integration modal wait for connection types to load instead of rendering the wrong monitor form or crashing.

Core

• Astralis App Coverage Expands: Many Admin UI pages are now available in the Astralis app:
  • Settings: Ontology, Custom fields, Roles, Users & roles, API credentials, the merged Organization settings page (Organization, Locations, Regulations), the merged DSR settings page (privacy request settings, storage, and pre-approval webhooks), Notification preferences (email and chat providers, messaging templates, and digests), and a Feature preview page.
  • Properties: the list, add, and edit pages, with Domains, Domain verification, and the Privacy Center form builder.
  • Consent: privacy notices, vendors, configuration settings and publisher restrictions, and the consent report and metrics dashboard.
  • Privacy requests: the Request Manager and the privacy request detail page.
  • App shell: navigation gains a breadcrumb dock and balanced rails, the light-mode background is lighter, the background animation runs at 30fps with a new static mode, feature-flagged navigation items are hidden to match the Admin UI, unmatched URLs render on an in-app 404 page, the context panel opens on hover only from the navigation rail, and placeholder copy is removed from Home.
• Multi-Organization Scoping: The multi-organization work continues, still inert for deployments not running Advanced RBAC with the multi-org feature flag. This slice covers:
  • User and role management: user management and role assignment are scoped to an admin's accessible organizations. The Roles tab shows organization-bound role grants and their inherited reach, and a new GET /plus/rbac/me/grants endpoint returns the caller's current role assignments with their permissions.
  • Consent writes: a scoped admin can no longer update or delete another organization's privacy notices, experience configurations, TCF configurations, or publisher restrictions.
  • Detection & Discovery: monitor creation and updates, identity provider monitor creation, bulk staged-resource actions, and task stop, retry, and dismiss are authorized per organization. The proposed-monitor database listing and the stewardship-inheritance recompute are scoped to the caller's organizations.
  • Reports: the data map report and export, asset reporting, the consent report, and the website monitor report only show data in organizations the user can access.
  • Messaging templates: messaging templates, including property-specific ones, belong to an organization, with label uniqueness per organization.
  • Organizations page: a newly created sub-organization appears in the hierarchy tree immediately, and Save returns to disabled after a successful save.
  • Includes database migrations.
• Smaller Access Tokens: Access tokens no longer repeat scopes that the user's roles already grant, so an owner login token now fits in a single browser cookie. Authorization is unchanged.
• Branded, Configurable Invite Email: The user invite email carries Ethyca branding and is an admin-configurable messaging template, and the accept-invite link always renders.
• Current User Endpoint: A new GET /api/v1/user/me endpoint returns the calling token's own user.
• Bedrock IAM Role Assumption: Bedrock LLM calls can assume an IAM role, configured through the new FIDESPLUS__BEDROCK__ settings. The Bedrock connection test accepts a role ARN on its own and reports the resolved role, the worker's identity, and the STS error code, and a new credential_source option pins which credential pathway the test uses. Role assumption and Bedrock access denials fail immediately instead of being retried.
• Telemetry Improvements: Every telemetry signal identifies the process that emitted it (webserver, worker, or Temporal worker, plus the Temporal task queue), and the assessment worker now exports telemetry. The Temporal metrics bridge publishes counters as counters rather than gauges, so rate and increase queries see every burst, and request tracing no longer emits duplicate per-pod HTTP server metrics.
• Security:
  • weasyprint bumped to 70.0.
  • The /health/workers endpoint no longer lets unauthenticated callers stall the webserver. The worker ping runs off the event loop, is bounded by the new FIDES__CELERY__WORKER_PING_TIMEOUT setting, and runs at most once at a time per process, with concurrent callers served the last result.
  • Deleting a discovery monitor requires the discovery_monitor:update scope rather than discovery_monitor:read.
• Bug Fixes:
  • Creating a taxonomy element whose key another taxonomy already uses returns an error naming that taxonomy, and deleting an element that still has children is refused with the child keys listed. Includes a database migration.
  • Assigning a role a user already holds returns a clear conflict error instead of a server error.
  • Astralis permission gates no longer fail closed after a page reload.
  • Single-line truncated text no longer overflows its container in description lists, and spacing utilities apply correctly to components in the Astralis app.

Database schema & data changes

• This release includes database migrations for assessment output (artifacts, findings, evidence, risks, and mitigations), framework content ratification and grounding verdicts, the organization-scoped assessments v2 gate decision ledger, policy document types, documents, and versions and their scopes, secrets manager configurations and their link from integrations, the asset_report:read scope, privacy request risk signal filtering, organization ownership of messaging templates, website monitor remote job tracking, cross-source catalog links and canonical groups, the staged-resource URN encoding backfill, and taxonomy key uniqueness.

Notes

• OAuth clients with an explicit datamap:read scope need asset_report:read. Roles are backfilled automatically, but OAuth clients created with an explicit datamap:read scope are not. Add asset_report:read to them with PUT /api/v1/oauth/client/{client_id}/scope before upgrading, or their calls to the asset report endpoints will return 403.
• Rollback is limited to 2.94.0 once the URN backfill runs. After the post-upgrade backfill re-keys existing monitors to the new staged-resource URN encoding, the deployment cannot be rolled back below 2.94.0.
• The privacy request CSV export omits identity and custom field columns unless asked. API callers of the CSV export must pass include_identities and include_custom_privacy_request_fields as true to keep the previous columns. The Admin UI already does and is unaffected.
• Pre-encoded database passwords must be stored literally. A stored database password is now sent to the driver exactly as stored. A deployment that worked around the old space-handling bug by storing a URL-encoded password, such as p%40ss for p@ss, must change it to the literal password or the connection will fail after upgrading.
• Polling access tasks write a pointer manifest by default. Set FIDES__EXECUTION__POLLING_ACCESS_MANIFEST_ENABLED=false to restore the previous concatenated write.