Skip to main content

Public roadmap

This page is a snapshot of where Control Zero is heading. It is updated roughly monthly and lags the engineering tracker by a release or two on purpose: we list things here once they are real enough that customers can plan around them.

We do not promise dates on this page. If a date matters to your rollout, the dashboard release notes and the changelog are the source of truth for what is actually shipped.

Shipped, Q1 to Q2 2026​

What has landed and is generally available.

  • Approvals (human-in-the-loop). Sensitive actions can pause for a human reviewer before they execute. End-to-end Python, Node, and Go SDK support. Dashboard inbox, per-approval drawer, and configurable timeouts and TTL grants. RBAC narrowing so only admins can edit policy or revoke grants.
  • Multi-project dashboards. The Governance Home, Analytics, and Audit pages all accept a project filter. Org admins can scope every view to a single project without leaving the org.
  • Stable, hashed argument fingerprints. Every SDK now produces an RFC 8785 canonical hash of the arguments passed to a guarded call. Policies can match on the hash without ever transmitting raw argument values.
  • Tamper response. Policy bundles are signed and encrypted; on tamper the SDK enters a configurable quarantine mode (warn, deny, deny-all). Configurable per org and per policy.
  • Approvals identity provenance. Every approval grant carries a verifiable chain of who requested, who approved, and which API key resolved the identity. Recorded in the immutable audit trail.
  • Policy library, republish flow, and detach as hard delete. Library policies can be versioned, republished, and detached cleanly with a distinct paused state for temporary disablement.
  • In-app invite inbox. Pending invites surface on a header bell even when the email got spam-trapped.
  • Browser extension (beta). Govern actions taken in claude.ai and chat.openai.com from the same policy bundle as your SDKs.
  • Shadow AI Scout (beta). Detect unsanctioned AI traffic on the network.
  • Compliance reports. SOC 2 style scoped reports across the immutable audit trail. (We do not claim certification on this page; ask your account contact for current certification status.)

In flight​

What we are actively building, with usable previews already in customers' hands.

  • Dual-write storage cutover for self-managed. Self-managed installs are migrating from one transactional store to another behind a feature flag, with a parity harness comparing every read and write. Goal: a single, well-understood store across SaaS and self-managed for everything except the analytics path.
  • Approvals: delegation and group approvers. A pending request can be delegated to another reviewer; a request can target a group rather than a single user. Identity and audit semantics already shipped; UI is in beta.
  • Approvals: notification channels. Slack, email, and webhook fan-out for pending requests, with per-channel templates and quiet hours.
  • Coverage gaps. A first-class view of which users, agents, and projects do not yet have any governed calls, so an admin can spot a missing rollout before it becomes a blind spot.
  • Onboarding flow refresh. Server-persisted four-step checklist, deep-linkable for support handoff, with cache-clear resistance so a returning user is never bounced back into onboarding.

Next up​

What we plan to start in the next quarter, scoped enough to commit to but not yet built.

  • Args fingerprinting, phase 1B: policy matching by hash. Phase 1A shipped the canonical hash. Phase 1B turns it into a policy primitive so you can deny by example without listing every variant.
  • Approvals: time-bound grants with budget. Approve once for the next N matching calls, or for the next N minutes, with a per-grant spend cap.
  • Tighter MCP server. Let an AI assistant manage your Control Zero policies through the same trust boundary as everything else. Today the MCP server is read-mostly; we are rolling out write actions one verb at a time.
  • Policy recipe library, expansion. More copy-paste playbooks for common scenarios. Submissions welcome.

Considering​

Ideas under exploration. No commitment that they will ship in this form, or at all.

  • Native policy preview in the dashboard. Edit a policy, see exactly which historical calls would have flipped from allow to deny, before publishing.
  • Args fingerprinting, phase 2: similarity match. Match on near-duplicates rather than exact hashes, for fuzzy denials.
  • Per-user adaptive thresholds. Approval cadence that learns from a user's confirmed-safe history.
  • Compliance pack: HIPAA report scaffolding. SOC 2 style scaffolding adapted to a HIPAA narrative.
  • Self-hosted, single-binary distribution. A reference deployment with no orchestrator dependency, for very small teams.

How to influence this list​

  • File a feature issue on the GitHub tracker with the feature label.
  • Email team@controlzero.ai for enterprise-shaped requests.
  • For security-shaped requests, see the security policy.

Last updated: 2026-05-18.