Changelog

Register named applications and group multiple API keys under each one. Audit-log and DLP-event records now carry the application's identity alongside the user, so governance reporting can answer which application sent a request, not just which key. Attribution is derived from the credential itself — the client-supplied identity header that could be set by the caller has been removed.

Compare and Summarize previously dispatched on Arbitex's platform credential even for organizations that had registered their own provider keys, which meant rotation, revocation, and primary-key designation did not apply to those two modes. All dispatch paths now resolve through a single credential seam, so your own key and its controls are honored in single, compare, and summarize alike.

The model selection list is now built from the providers your organization has actually connected and synced, instead of falling back to a fixed list. Organizations running entirely on their own provider credentials no longer see models their keys cannot reach.

Test a draft policy or DLP rule against sample content and see exactly what the deployed enforcement engine would do with it. Dry-run now executes on the real engine rather than an in-browser approximation, so rules that reference built-in detectors return the same verdict in testing as they will in production.

The tamper-evident audit chain now emits periodic cryptographically signed checkpoints, with the signing key held in a managed key vault. Auditors can verify a span of audit history against a signed checkpoint instead of replaying the entire chain, and the signature is verifiable independently of Arbitex.

Administrators can now set multi-factor authentication to off, optional, or required for the whole organization, with an enrollment guard that brings existing users into compliance on next sign-in. Every enrollment and enforcement change is recorded in the audit log.

Organizations can author their own content-category keyword and negative-keyword lists that extend or override the built-in taxonomy. Custom lists are scoped to your organization only and are evaluated by the same policy conditions as the built-in categories.

A scan could report a Block verdict in a case where overlap resolution had already discarded every match that justified blocking, leaving the reported reason inconsistent with the action taken. Overlap resolution, action selection, and confidence are now computed at a single decision point per span, so the recorded verdict is always the one that was enforced.

An error partway through a File Inspector attachment scan could be caught and recorded as a completed scan with no risk found — a control that silently did not run while reporting clean. Scan infrastructure errors now route to the organization's configured scan-error action and always reach the audit feed.

Retention periods could be set on an Outpost deployment but nothing applied them: there was no purge job and the write-ahead log was never compacted, so audit storage grew without bound and expired records were never released. A retention sweep now enforces the shortest applicable period per data class and compacts the log after each purge.

Every authoring surface in the console now works as a changeset. Edits are staged rather than applied immediately, presented as a reviewable diff, checked for conflicts against concurrent changes, and committed only when you explicitly apply them. Each applied changeset is retained in history, giving change control over security policy the same shape auditors expect elsewhere in the stack.

Certificate renewal previously expected a centrally generated private key to be delivered to the Outpost over the network. That path has been removed outright. The Outpost now generates its own key locally and sends only a certificate signing request; the control plane returns a signed certificate and never sees the key. No setting restores the old behavior.

Users can enroll a time-based authenticator or a passkey themselves from the profile menu, without an administrator provisioning it for them. Backup codes are issued at enrollment and every enrollment event is audited.

Administrators can verify sending domains and configure SMTP relay settings for the Email Relay channel directly in the console, then validate the configuration with a connection test before depending on it in production.

Generate and download an actual evidence report for a compliance framework from the console, rather than only toggling a bundle on. Reports are produced against your organization's own enforcement records and are retained for later retrieval.