Changelog

The platform is now versioned and available as the Arbitex 2026.1 private beta, consolidating this quarter's channel, policy, and deployment work behind a single release line.

A visual console for building DLP policy. Apply prebuilt compliance policy packs and reorder the rule chain with drag-and-drop — administrators can see exactly which rules run, and in what order, before any change reaches live traffic.

Test policy and DLP rules against simulated user groups before deploying them. Preview exactly what a rule would do across different teams and roles, then adjust with confidence before it ever touches production traffic.

Organizations now control which content modalities are permitted. Text and tool-use are enabled by default; image, audio, and realtime modalities are admin-enabled per modality and enforced at the API so unapproved content types never reach a model.

Administrators can now configure the platform end to end from the control plane — security and DLP, system settings, monitoring and users, models and routing, and budget controls are all wired through to enforcement.

DLP inspection now extends to email. Messages routed through the SMTP relay pass through the same multi-layer content inspection pipeline that protects API and chat traffic, with block and redaction actions enforced end to end.

Block and redaction actions are now reliably enforced across every channel and every tier of the inspection pipeline. If inspection cannot complete, requests fail closed rather than passing sensitive content through.

The tamper-evident audit log has been hardened for sustained high-volume traffic, preserving chain integrity and ordering guarantees even under heavy concurrent write pressure.

Container images are now cryptographically signed and verified before deployment, workloads authenticate with short-lived workload identity instead of long-lived secrets, and dependencies are continuously monitored for newly disclosed vulnerabilities.

Users can attach files to AI conversations. Every attachment passes through the DLP pipeline before reaching any model — documents are scanned for sensitive content, quarantined if policy triggers, and linked to the originating message for full audit trail.

Administrators can review, release, or permanently delete quarantined email messages from a dedicated dashboard. Quarantine stats, filtering, and pagination provide visibility into email DLP enforcement across the organization.

Model Context Protocol servers are now governed by the Arbitex policy engine. Per-server security policies, a dedicated security API, and admin controls let organizations manage which MCP servers can connect, what data they can access, and how interactions are logged.

Token consumption is now attributed per provider, per department, and per user. Budget caps and usage quotas enforce spending limits in real time, with alerts before limits are reached. Cost analytics dashboards give finance and engineering shared visibility into AI spend.

Three new compliance packs map detection rules and enforcement actions to the EU AI Act, NIST AI Risk Management Framework, and ISO/IEC 42001 AI management system standard. Activate a pack and governance follows automatically.

AI conversations are now automatically titled based on content. A new auto_titled flag tracks whether the title was generated or manually set, and a regenerate endpoint lets users request a fresh title at any time.