Skip to content
Trust Center

Security you can verify before you deploy.

We are asking to run a sensor on your fleet and to help you govern AI risk. That only works if you can see exactly how we handle data, what we collect, what we don't, and where we stand on compliance. This page is the single source of truth — and we tell you what's live, what's in progress, and what's on the roadmap. Honestly.

Last updated: August 14, 2026

Compliance status

We publish our real status rather than badges we haven't earned. Where a certification is not yet complete, we say so — and we can share our current controls, policies and evidence under NDA today.

FrameworkStatusNotes
SOC 2 Type IIn progressControls implemented; readiness assessment underway — Type I attestation targeted Q1 2027.
SOC 2 Type IIPlannedObservation period to follow Type I.
ISO/IEC 27001RoadmapISMS controls mapped; certification targeted next.
ISO/IEC 42001 (AI management)AlignedOur own product maps findings to 42001; internal alignment underway.
GDPRAlignedDPA available; data-subject request process in place. See Privacy.
EU AI ActMappedFindings map to Art. 9 / 15 / 55 obligations; we deliver the evidence structure.
Under evaluation? Request our security package — current SOC 2 readiness report, policies and architecture under NDA (an internal penetration test has been completed and an independent third-party review is underway; the summary will be available under NDA). See how we handle the privacy of monitored employees.

What the sensor collects — and what it never does

What it collects

Metadata of AI usage: process/binary names, MCP config locations, API-key presence and provider (never the value), installed AI CLIs/SDKs, browser-extension IDs, and egress destinations to known AI endpoints — attributed to machine and user.

What it never collects

Not the contents of your files, not secret values (masked at the point of collection), not keystrokes, not prompt/response bodies, not personal browsing. Command lines are scrubbed of secret-shaped tokens before they leave the endpoint.

Architecture & data residency

  • Sensor is read-only. It observes and reports. It does not modify files, kill processes, or change security settings.
  • Enforcement is a separate, explicitly-authorised channel. Response runs through a distinct policy-and-response path an operator opts into — the content-DLP policy engine and remote-pause today, with host/agent enforcement on the roadmap — never the sensor by itself. This is the boundary architects ask about, stated plainly.
  • Deployment options. Managed (our AWS, single-tenant isolation) or fully in-VPC — the console and the classification model run inside your AWS account with no data egress. In-VPC uses Amazon Bedrock for reasoning so prompts never leave your boundary.
  • Data residency. Managed tenants can be pinned to a region (US or EU). In-VPC keeps 100% of data in your account.

Encryption, access & keys

  • In transit: TLS 1.2+ everywhere; HSTS; sensor-to-console is mutually authenticated with a per-device enrollment secret.
  • At rest: encrypted storage; secrets held in AWS SSM SecureString / your KMS.
  • Access control: role-based access across four roles — Viewer / Analyst / Admin / Owner (plus a Platform Admin role for Araghatta staff) and custom roles — TOTP MFA on every account, and a tamper-evident audit trail on every action. SSO (OIDC) and SCIM 2.0 are live (Enterprise tier); SAML 2.0 is on the roadmap.
  • Integration secrets (Slack webhooks, Jira tokens) you enter are encrypted at rest and never returned to the browser.

Subprocessors

We keep this list short and current. Managed deployments use:

SubprocessorPurposeRegion
Amazon Web ServicesHosting, storage, secrets (KMS/SSM)US / EU (per tenant)
Amazon SESTransactional email (enquiries, alerts)US
Reasoning backend (configurable)AI triage — hosted API, or Amazon Bedrock in-VPC, or fully localPer configuration

In-VPC deployments have zero external subprocessors for data processing. See the canonical, dated sub-processor list; a signed version is part of the DPA.

Vulnerability disclosure & incident response

  • Report a vulnerability: info@araghatta.com. We acknowledge within 2 business days and coordinate disclosure in good faith.
  • Incident response: defined runbook; customer notification for confirmed incidents affecting your data within 72 hours, with a root-cause summary to follow.
  • Product security: the sensor binary is versioned and signed; updates are pinned by version and verified before rollout.