Home / Roadmap
Planned work. Not available today.
Each row names something we plan to build, next to what the product does today. Items on this page are planned, not available, and are not commitments or dates. Capabilities that are available today are on Platform status and in the capability registry.
How to read this page
Planned means we intend to build it and it is not available. In progress or In validation means work has started, and it is still not available. Do not buy on the basis of a roadmap item: if a feature matters to your decision, ask us for its current state in writing. Published 2026-10-16.
Roadmap
Endpoint sensor
What the sensor does today is on the sensor page. These are the planned extensions.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Windows editions and server roles, validated | Windows endpoint and server collectors are built and available. They have not yet been run at scale on production hardware. | Validated coverage of Windows 10 and 11 Enterprise, Pro and LTSC and Windows Server 2016 to 2025, including Server Core, domain controllers, RDS and failover clusters, with an MSI package and Group Policy deployment. | Planned |
| Kernel-level network containment | Quarantine uses each operating system's own firewall. The sensor re-checks it every 20 seconds, puts it back if it is changed or switched off, and re-applies the last policy it received when it starts. | Network filtering built into each operating system's network stack, so a local administrator cannot remove containment and it applies from early boot, with rules per program. The macOS and Windows parts need approvals from Apple and Microsoft. | Planned |
| Real-time tracing without extra dependencies, and on macOS and Windows | Live tracing of programs and connections is available on Linux. macOS and Windows collect on a schedule. | A built-in tracer with no extra dependency, validated on bare metal, and event-assisted tracing on macOS and Windows. | Planned |
| Process-attributed DNS queries and SNI | Connection details for AI services are attributed to the program on the machine. DNS evidence arrives through the proxy and DNS connectors. | DNS query capture and SNI attribution on the endpoint, tied to the process. | Planned |
| Cloud metadata read from the host on Google Cloud and Azure | On AWS servers the sensor reports the posture of the cloud metadata service and the attached role. For Google Cloud and Azure, identity and reach come from the read-only connectors. | Service-account metadata from Compute Engine and managed-identity metadata from Azure VMs, reported by the sensor itself. | Planned |
| Kubernetes admission-controller and policy integration | A read-only Kubernetes node sensor and a read-only cluster connector. Recommendations only; nothing is enforced at admission. | An admission-controller integration that applies approved-AI policy to workloads at admission. | Planned |
| Validated Linux distribution matrix | Tested end to end in containers on Ubuntu, Debian, RHEL and its clones, Amazon Linux, SUSE, Fedora and Alpine, on x86-64 and arm64, and in a virtual machine on Ubuntu. | Validation on virtual machines and bare metal for each distribution, including the packaged service and SELinux, and RPM packages. | Planned |
| Deployment kits for more tools: live validation | Deployment guides for Intune and Jamf are available. Kits and step-by-step guides for Kandji, Ansible and Configuration Manager are written; they have not yet been run on live tools. | Validate each kit on a pilot fleet of its tool and publish the guides. | In validation |
Roadmap
Cloud posture and validation
The AWS, Azure and Google Cloud connectors and the agent red-team read and test what their pages describe today. These are planned depth and coverage.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Oracle Cloud posture beyond AI-service discovery | An Oracle Cloud connector (beta) discovers AI services. It does not read access policies or storage posture. | Sensor-led posture on OCI instances (metadata and instance principals), tenancy IAM policy and bucket posture, and Cloud Guard findings ingestion. | Planned |
| AWS effective access with conditions, Deny, SCPs and resource policies | Effective access is an upper bound that errs on the side of flagging. | Evaluate conditions, Deny statements, permission boundaries, service control policies and resource policies, so the result narrows from an upper bound toward the real access. | Planned |
| Azure depth: deny assignments, PIM, server context, network paths | Virtual machines, managed identities, role assignments, storage accounts and AI services are covered. | Deny assignments, management-group scope, PIM eligibility, scale-set, AKS and Arc server context, SAS policy and immutability checks, and NSG-based reverse hybrid paths. | Planned |
| Google Cloud depth: GKE, Cloud Run, organization policy | Virtual machines with attached service accounts, project access, storage and enabled services are covered. | GKE and Cloud Run workload context, Workload Identity posture, organization policy constraints, key inventory, and bucket logging and retention checks. | Planned |
| Cloud-provider findings as corroborating evidence | The attack-path engine uses the read-only connector data only. | Ingest AWS CloudTrail, Config, Security Hub, GuardDuty and Security Lake findings, Azure Activity Logs, Defender for Cloud and Sentinel incidents, and Google Cloud Audit Logs, Security Command Center and SecOps detections as corroboration. | Planned |
| Resolve an exposed key to the identity it grants | The forward path shows the worst reach of any identity in the connected account. The exposed key is never used. | With the customer's authorization, resolve an exposed key to its owning identity and show its actual effective permissions. | Planned |
Roadmap
Identity, approvals and access
SSO, SCIM and role-based access work today. These are the planned controls around them.
| Item | Available today | Planned | Status |
|---|---|---|---|
| SCIM offboarding that revokes AI access | Deactivating a user in your identity provider disables the account and revokes that user's enrolled sensors. | Trigger an AI-access review and revocation (AI tools, provider keys, agent identities) for the offboarded user. | Planned |
| Approval step for more response actions | Per-host quarantine approval can require a ticket or change reference, a time limit and a second administrator, set by an Owner. An approval that has ended reads as not approved, so the host returns to automatic protection. Catalog requests, vendor access and access reviews already need a different approver. Temporary Admin access needs a reason and a different Owner's approval. | The same approval step for other response actions: enforcement mode changes, connector actions and policy changes, with target scope, expiry and a separation-of-duties rule on each. | Planned |
Roadmap
Connectors, integrations and data handling
Connectors are read-only and scheduled today. These are planned additions to how data comes in and goes out.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Native SIEM connectors: Chronicle and Security Lake, and live validation | Findings flow to Splunk and QRadar directly, and as signed messages that feed Sentinel, Elastic and Chronicle through your own pipeline. Native QRadar LEEF, Elastic ECS and Microsoft Sentinel (Logs Ingestion API) outputs are built; they have not yet been run against live systems. | Validate the QRadar, Elastic and Sentinel outputs against live systems, and add a Google Chronicle ingestion output and an Amazon Security Lake writer. | In validation |
| Hosted remote MCP endpoint | A local assistant connection with scoped, revocable keys. Web-based assistants cannot connect to it today. | A hosted remote endpoint with the same scoped, read-only, audited key model for web-based assistants. | Planned |
| Vendor-certified EDR and ITSM integrations, packaged ServiceNow app | Coverage reconciliation with CrowdStrike and Defender, and ServiceNow tickets, read-back and owner sync, configured in the app. Not vendor-certified. | Vendor certification of the two-way integrations and a packaged ServiceNow connector. | Planned |
| More ownership sync sources | Ownership sync from Okta, Microsoft Entra ID and ServiceNow, plus manual entry and CSV import. | Intune and Jamf as MDM sources, CMDB sources, and CrowdStrike and Defender as endpoint sources. | Planned |
| Response actions through connectors and traces | Connectors are read-only. Response today is alerts, tickets, sensor revocation and the Linux enforcement sensor. | Connector-driven revoke, block and quarantine with dry-run, approval, a scoped service role, audit and rollback; live response actions from execution traces; and coordinated response across endpoint, cloud and identity. | Planned |
| Inline cloud-side policy enforcement | Cloud connectors are read-only, and the change-risk gate is advisory. | Apply an approved policy at the cloud control plane, with the same simulate, approve and rollback steps as endpoint enforcement. | Planned |
| Event-driven connector collection | Connectors collect on a schedule, plus push endpoints for log sources. | Event streams where the vendor offers them, so a change appears without waiting for the next poll. | Planned |
| Azure OpenAI and Vertex AI reasoning backends | Reasoning can run on a hosted model or inside your own cloud account, or locally. Scoring is always deterministic. | Azure OpenAI and Google Vertex AI as customer-selected reasoning backends. | Planned |
Roadmap
Governance and compliance
Framework mapping, the risk register and Approved AI work today. These are planned extensions.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Automatic EU AI Act mapping for more articles | Findings map automatically to the articles Sargha covers. Other articles are referenced by hand. | Automatic mapping and evidence status for Art. 6, 9, 11, 12, 26, 50, 55 and 72. | Planned |
| Safety and Security Model Report generated in the product | Delivered as an engagement report (see Services). The product produces signed per-framework compliance reports that anyone can verify. | A product-generated, signed Safety and Security Model Report populated from live evidence. | Planned |
| Shadow AI Risk Index in the product | The posture score is a per-host number weighted by severity. The Shadow AI Risk Index is a published framework (see Research). | Offer the five-surface Index as a second score beside the posture score. | Planned |
Roadmap
Enterprise and deployment
Managed hosting is AWS us-east-1 today. These are planned deployment, resilience and scale items.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Regional data planes, EU-region tenants and data-residency pinning | Managed tenants run in one US region today. Customer-cloud deployments keep telemetry in your own account and region. | Regional data planes that ingest and store telemetry inside the selected boundary, EU-region managed tenants, and per-tenant region choice with residency pinning. | Planned |
| Customer-managed and dedicated tenant keys on managed tenants | Managed tenants use service-managed keys. Dedicated and customer-managed keys are offered by arrangement, as custom work. In a customer-cloud deployment you use your own keys. | Customer-managed keys (your KMS or HSM) as a standard, packaged option for managed tenants. | Planned |
| Private ingress and egress | Secure connections to the collector, with proxy support. Private connectivity controls are offered by arrangement, as custom work. | Private connectivity (AWS PrivateLink and Google Private Service Connect style endpoints) for sensors, connectors and exports, as a standard option. | Planned |
| Dedicated managed tenant as a packaged option | Available by arrangement as custom work. | A packaged dedicated tenant with a standard service description. | Planned |
| Air-gapped and zero-egress deployment | Customer-cloud deployment is available, with reasoning inside your account or on a local model. | A documented air-gapped package with local reasoning and no external email relay. | Planned |
| Multi-zone application tier, durable queues and failover | Automatic recovery, a resilient managed database and health monitoring are in place today. | A multi-zone application tier, durable ingest queues, health-based failover and independently recoverable tenant storage. | Planned |
| Published recovery objectives and scheduled restore exercises | Point-in-time database recovery is in place. Published recovery objectives and restore drills are planned. | Documented RTO and RPO by deployment tier, scheduled restore exercises and customer-visible recovery evidence. | Planned |
| 100k+ endpoint fleet scale, load-tested | Large-fleet design is in place. Fleet-scale load testing has not been published. | Load tests at 10,000 and 100,000 hosts per tenant, and published sizing envelopes. | Planned |
Roadmap
Product workflow
Fix as code, decisions and reports work today. These are planned improvements.
| Item | Available today | Planned | Status |
|---|---|---|---|
| Credential rotation that calls the provider | A step-by-step script guides rotation. You create and revoke the key at the provider. | Scripts that call the provider's API to rotate a key and verify the old one is revoked. | Planned |
| Complete DNS, proxy and cloud deny policies | Starter block lists and an explicit-deny cloud policy are provided. On the endpoint, a denied site and everything under it are blocked on Windows, macOS and Linux and, as an organization option, the addresses behind them are dropped. | Complete rule sets, plus Azure and Google Cloud deny policies. | Planned |
| Role views scoped by permission | Four tabs, one per job. Tabs arrange the view; access is still set by role. | Permission-scoped role views, so a tab shows only what that role may see. | Planned |
Roadmap
Assurance and documentation
Items in progress. None of these is attained today.
| Item | Available today | Planned | Status |
|---|---|---|---|
| SOC 2, ISO/IEC 27001 and CSA STAR | All three are in progress. None is attested or certified. | Complete the audits and publish the reports and registrations here when they exist. | In progress |
| Independent third-party penetration test | An unauthenticated web assessment (DAST) of the public site was run internally in August 2026 (summary under NDA). An independent third-party penetration test, including the sensor and the collector, is being commissioned; it has not been performed yet. | Publish the executive summary in the Trust Center. | In progress |
| Self-service documentation portal | Deployment guides and reports are shared on request; the sample reports are on this site. | A self-service documentation portal (planned host: docs.araghatta.com). | Planned |
Get started
See the AI on your fleet in about a week.
Run the sensor across your estate and receive an AI risk report and an AI Bill of Materials. The assessment is free and carries no commitment.